A framework of intelligent decision support system for ...

3 downloads 188 Views 882KB Size Report
Support System for police that can work as a decision tool. The same is .... Police IT in Karnataka, Thana Criminal Tracking System (TCTS) in West Bengal,.
Journal of Enterprise Information Management A framework of intelligent decision support system for Indian police Manish Gupta B. Chandra M.P. Gupta

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

Article information: To cite this document: Manish Gupta B. Chandra M.P. Gupta , (2014),"A framework of intelligent decision support system for Indian police", Journal of Enterprise Information Management, Vol. 27 Iss 5 pp. 512 - 540 Permanent link to this document: http://dx.doi.org/10.1108/JEIM-10-2012-0073 Downloaded on: 22 September 2014, At: 10:23 (PT) References: this document contains references to 31 other documents. To copy this document: [email protected] The fulltext of this document has been downloaded 20 times since 2014*

Users who downloaded this article also downloaded: Richard William Adderley, Peter Musgrove, (2001),"Police crime recording and investigation systems – A user’s view", Policing: An International Journal of Police Strategies & Management, Vol. 24 Iss 1 pp. 100-114 Kenneth J. Novak, Jennifer L. Hartman, Alexander M. Holsinger, Michael G. Turner, (1999),"The effects of aggressive policing of disorder on serious crime", Policing: An International Journal of Police Strategies & Management, Vol. 22 Iss 2 pp. 171-194 Brandon R. Kooi, (2009),"Problem#oriented Policing and Crime Prevention20091A. Braga, . Problem# oriented Policing and Crime Prevention. Monsey, NY: Criminal Justice Press 2008. , ISBN: 13: 978#1# 881798#78#1", Policing: An International Journal of Police Strategies & Management, Vol. 32 Iss 4 pp. 806-808

Access to this document was granted through an Emerald subscription provided by 427157 []

For Authors If you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service information about how to choose which publication to write for and submission guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.

About Emerald www.emeraldinsight.com Emerald is a global publisher linking research and practice to the benefit of society. The company manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as providing an extensive range of online products and additional customer resources and services. Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive preservation. *Related content and download information correct at time of download.

The current issue and full text archive of this journal is available at www.emeraldinsight.com/1741-0398.htm

JEIM 27,5

A framework of intelligent decision support system for Indian police

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

512 Received 6 October 2012 Revised 8 March 2013 13 April 2013 Accepted 14 April 2013

Manish Gupta and B. Chandra Department of Mathematics, Indian Institute of Technology Delhi, New Delhi, India, and

M.P. Gupta Department of Management Studies, Indian Institute of Technology Delhi, New Delhi, India Abstract Purpose – The purpose of this paper is to introduce architecture of an Intelligent Decision Support System to fulfill the emerging responsibilities of law enforcement agencies. Design/methodology/approach – The proposed Intelligent Police System (IPS) is designed to meet the emerging requirements and provide information at all levels of decision making by introducing a multi-level structure of user interface and crime analysis model. The proposed framework of IPS is based on data mining and performance measurement techniques to extract useful information like crime hot spots, predict crime trends and rank police administration units on the basis of crime prevention measures. Findings – IPS has been implemented on actual Indian crime data provided by National Crime Records Bureau (NCRB), which illustrates effectiveness and usefulness of the proposed system. IPS can play a vital role in improving outcome in the crime investigation, criminal detection and other major areas of functioning of police organization by analyzing the crime data and sharing of the information. Research limitations/implications – The research in intelligent police information system can be enhanced with some important additional features which include web-base management system, geographical information system, mobile adhoc network technology, etc. Practical implications – IPS can easily be applied to any police system in the world and can equally be useful for any law enforcement agencies for carrying out homeland security effectively. Originality/value – The research reported in this manuscript is outcome of the research project funded by NCRB. This paper is the first attempt to build framework of IPS for Indian police who deal with large volume and high rate of crimes that are unmatched to any police force of the world. Keywords Data envelopment analysis, Crime data mining, Intelligent decision support system, Police information system Paper type Research paper

1. Introduction After the September 11, 2001 terrorist attacks on the World Trade Centre in New York City and the Pentagon, most of the governments initiated efforts to modernize law enforcement authorities’ intelligence collection and processing capabilities (Bella et al., 2010; Chen et al., 2003; Chung et al., 2006; Gupta and Mitra, 2005; Krishnamorthy, 2003). It also saw increasing global cooperation among policing units of various countries. Journal of Enterprise Information Management Vol. 27 No. 5, 2014 pp. 512-540 r Emerald Group Publishing Limited 1741-0398 DOI 10.1108/JEIM-10-2012-0073

This paper is an outcome of a project funded by National Crime Records Bureau (NCRB), New Delhi. The authors are thankful to the Director NCRB and the team for sharing valuable knowledge in this field. The authors are also thankful to anonymous reviewers for their valuable comments which were found helpful in improving the manuscript.

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

The major challenge before police organizations was and still is to handle large number of information and huge volume of records pertaining to crime and criminals. Intelligent support systems can play a vital role in improving outcome in the crime investigation, criminal detection and other major areas of functioning of police organization by facilitating recording, retrieval analysis and sharing of the information. Highly advanced analytical methods are required to extract useful information from the large amount of crime data. Data mining (Han and Kamber, 2006) is looked upon as a solution to such problems. In the past, data mining has emerged as a promising technique for crime detection, clustering of crime locations, criminal profiling, predictions of crime trends (Boba, 2001; Chung et al., 2006; Gupta et al., 2007; Levine, 2002) and many other related applications (Oatleya and Brain, 2003). In particular, clustering (Han and Kamber, 2006; Jain et al., 1999), classification (Breiman et al., 1984; Haykin, 2008) and association rule mining (ARM) (Agrawal et al., 1993) are found very useful in dealing with crime-related data. Feature selection (Dy and Brodley, 2004) forms one of the foremost pre-processing techniques required for identifying the important features prior to applying data mining. Several techniques for feature selection (Dy and Brodley, 2004) have been employed in the past. In the context of India, efforts of the authors worth citing here where classification, clustering, semi-supervised clustering (SSC) (Bilenko et al., 2004) and multivariate time series (MTS) clustering (Chandra et al., 2008) in combination with dynamic data envelopment analysis (DEA) (Charnes et al., 1978; Tone and Tsutsui, 2010) have been effectively applied over large set of longitudinal crime data provided by National Crime Records Bureau (NCRB) (which is the prime organization for collecting Indian crime data), results of which are found very useful for the Indian police to meet and fulfill its emerging responsibilities and tasks of the police (Chandra et al., 2007; Gupta et al., 2007, 2008; Gupta and Mitra, 2005; Mitra and Gupta, 2008). What now needed is to integrate them together into a larger framework of Intelligent Support System for police that can work as a decision tool. The same is attempted here in this paper. A significant part of past literature confine to dealing with the Decision Support System (DSS) in traditional sense (Shim et al., 2002; Turban, 1995), some attempting a unified architecture for DSS (Teng et al., 1988; Janssen and Cresswell, 2005) and other multi-criteria DSSs (Siskos and Spyridakos, 1999). Researchers are now attempting a new generation of DSS, i.e. Intelligent Decision Support Systems (IDSS) (Bhargava et al., 2007), which demonstrated a need for combining data mining with multi-criteria decision making and also allow dynamic interaction of knowledge-based systems to address the major challenges of effective information management systems (Cheung et al., 2005; Peng et al., 2011). This serves the dual purpose of creating an automatic decision process or problem-solving process (replace human tasks) and supporting the end-users in their decision process (Shim et al., 2002; Weerakkodya et al., 2012). Primary focus of this paper is to introduce architecture of an Intelligent Police System (IPS), a kind of IDSS based on data mining and performance measurement techniques to fulfill the emerging responsibilities and massive task of Indian police who deal with large volume and high rate of crimes that are unmatched to any police force of the world. In the past, there have been many cases of terrorism in India, for example, Parliament Attack, 2001, Bombay Terrorist Attack, 2008, etc. that emphasize the need of an IPS which can counter such threats proactively and help the police to disseminate intelligence information effectively. The proposed IPS is an attempt in this direction. Here, feature selection algorithms have been used to extract

A framework of intelligent decision support system 513

JEIM 27,5

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

514

significant features from the huge volume of Indian crime data provided by NCRB. Classification techniques have been applied to identify crime patterns in the Indian crime data. Clustering and SSC methods have been applied to find crime hot spots in India. MTS clustering in combination with dynamic DEA has also been carried out on periodic crime data. MTS clustering forms homogeneous groups of Police Administration Units (PAUs) with similar crime trends prior to applying dynamic DEA which ranks PAUs on the basis of their effective enforcement of crime prevention measures. Details of the proposed architecture of IPS have been described in Section 4 of the paper. Section 2 briefly describes the structure and role of the Indian police in order to understand the Indian police system and its requirements. Section 3 presents the existing police information system in India. The need of IPS and its architecture along with functionality of its various sub-systems has been elaborated in Section 4. Section 5 illustrates the implementation of IPS prototype based on developed efficient techniques on Indian crime records. Concluding remarks comprise the last section of the paper. 2. About Indian police As per the constitution of India, police is a subject allocated to State. However, the Union Government have raised its own police forces called Para Military Forces for specific purposes. This consisted of Central Reserve Police Force for counter-insurgency operations, Border Security Force for guarding the international border, Indo-Tibetan Border Police guarding Indo-Tibet Border, Central Industrial Security Force for security of vital installations and government buildings and Special Security Bureau deployed on Indo-Nepal Border apart from undertaking the tasks of preparing border population to counter subversive activities from across the border. Given their unique nature of duties, Para Military Forces have very limited interface with common people (Mitra and Gupta, 2008). Police at the States level has maximum public interface as they are dedicated in providing public security, maintaining law and order and providing all other policing assistance and services to the people in general. Its highly hierarchical organization with 13 ranks in chain of command headed by the Director General of Police (DGP) as shown in Figure 1. Among these, the senior officers are drawn, by and large, from The Indian Police Service who do the supervisory work, the “upper subordinates” (inspectors, sub-inspectors and assistant sub-inspectors) who work generally at the police station level and the police constabulary who are delegated the patrolling, surveillance, guard duties and law and order work. The constabulary accounts for almost 88 percent of total police strength (Figure 2). Functionally, a Police Station is the basic unit of police administration (at the bottom) through which both crime (as enunciated in the Indian Police Act, 1861) and non-crime duties are discharged; complaints and First Information Reports (FIRs) are lodged (second from the bottom in the police hierarchy as portrayed in Figure 1). Police stations are manned by junior officers such as Inspectors, Sub Inspectors, Assistant Sub Inspectors, Head Constables and Constables who directly interface with people as well as main policing jobs like crime prevention, crime investigation, law and order, in collecting evidences, in preparing prosecution papers, etc. A district may have many Police Stations. District Police Chiefs (Senior Superintendent of Police (SSP) and Superintendent of Police (SP)) are responsible for supervision and monitoring of the works of police stations. SSP/SP is assisted by an ASP and few DSPs.

Director General of Police (DGP) [In charge of State Police] Additional Director General of Police (A DGP) Inspector General of Police (IGP) [Zone, collection of Ranges]

A framework of intelligent decision support system

Deputy Inspector General of Police (DIGP) [a Range, collection of Districts] Senior Superintendent of Police (SSP) [In charge of a bigger District]

515

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

Superintendent of Police (SP) [In charge of the District] Additional Superintendent of Police (ASP) Assistant/Deputy Superintendent of Police (ASP/DSP) [In charge of a Sub-division in the District] Inspector of Police [In charge of a Police Station] Sub Inspector of Police (SI) [In charge of a smaller Police Station] Assistant Sub-Inspector of Police (ASI) [Staff of the Police Station] Police Head Constable (HC) [Staff of the Police Station] Police Constable [Staff of the Police Station]

Figure 1. Police hierarchy

Level-I: DSP to DGP (0.88%) Level-II: Inspectors, SI and ASI (11.51%) Level-III: Police Constabulary (87.61%)

There has been significant growth in police manpower since independence (1947). On January 1, 2003, the total strength of the State Police Forces was 1,468,776 out of which civil police constituted 1,120,167 and the armed police 348,609. Percentage wise, the civil police accounts for 76.27 percent and armed police accounts for 23.73 percent. During the period 1947-2003, the police strength registered an increase of 280.5 percent. The civil police increased by 351.9 percent, the armed police by 161 percent. Improving functioning of the police stations, enhancement of skill and behavior of the police personnel at the cutting edge level have remained dominant strands of efforts of both the Central and State governments (Mitra and Gupta, 2008). “Police Reforms” is an important component of Government’s agenda on administrative reforms. Way back in 1970, the National Police Commission (NPC, also known as Dharamvira Commission) was set up to make a comprehensive review of police system at the national level. Since then, a number of Committee and Commissions specifically set up to go into the various aspects of “police reforms.” These included central

Figure 2. Indian police pyramid

JEIM 27,5

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

516

Committees headed by J. Ribeiro, a retired police chief (1998), K. Padmanabhaiah, Former Union Home Secretary (2000), Group of Ministers on National Security (2000-2001) and Justice Malimath, a retired Chief Justice and an eminent prison reformist (2001-2003), who all endorsed the recommendations of the NPC, with minor variations. These are testimony of government concern and response to increasing police role in public life, apart from prevention of crime and related works, to maintain social order and protect the individual freedom and privacy, etc. As the roles of the Indian police increases in each dimension of law enforcement, the existing police information systems are also going through major upheaval to meet such requirements of the Indian police forces. 3. Evolution of information systems in Indian police Computerization in Indian Police dates back to 1986 when NCRB was created with the aim to help the investigating agencies by providing them with extensive and updated information on crime and criminal data at State, National and International level such as modus operandi, personal data, finger print, photograph, criminal history and details of property which may be subject matter of crime. This led to development of islands of police information systems covering different subjects such as Fingerprint Analysis and Criminal Tracking System (FACTS), Motor Vehicle Coordination System, Talash Information System, Portrait Building System, etc. At the state level, several initiatives have been taken like, e-Cops in Andhra Pradesh, Police IT in Karnataka, Thana Criminal Tracking System (TCTS) in West Bengal, CAARUS in Tamil Nadu and HD-IITS in Gujarat. Figure 3 shows the evolution of Indian police information systems. The most notable work of NCRB is the Crime Criminal Information System (CCIS) which is developed as a national project of sharable database on crime and criminals at district, state and national levels for assisting investigations and supervising officers and police planners to formulate crime-control strategies. Figure 4 shows snapshot of the CCIS database, which displays a tabular view of FIRs among other tables. The FIR table contains records of every committed crime. CCIS is perhaps one of the biggest police applications in the world implemented in 35 states and union territories (UTs), 727 police districts and at the national level (http://ncrb.nic.in/ccis.htm). Most of the state police headquarters and district headquarters are covered by CCIS and so are some of the 14,000 þ police stations in the country. In 2005, it was upgraded as multi-lingual application with facility for different local languages, i.e. Marathi, Gujarati, Tamil, Kannada and Gurmukhi besides English and Hindi. The main objective of the CCIS is to computerize crime and criminal information collected by the Inquiry Officer, to link a crime to criminal and property, generate vital reports from the database and reduce manual effort and increase efficiency of the police. CCIS database is used for crime monitoring by monitoring agencies such as NCRB, State Crime Records Bureaus and District Crime Records Bureaus and to facilitate statistical analysis of crime and criminals related information

Figure 3. Evolution of computerization efforts in Indian police

CCIS (2001)

CIPA (2005)

CCTNS (2009)

NATGRID (2011)

A framework of intelligent decision support system

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

517

Figure 4. CCIS database snapshots

with the states and monitoring agencies. CCIS data are used for publishing online reports such as missing persons report and is also used as the basis for online query facilities that are available through the NCRB web site. In addition, it is also used by NCRB to publish an annual nation-wide Crime Report. The major limitation in CCIS is that the information is collected at the district level and not at the primary source of information which is the police station. It is a somewhat “standalone” system with no analytical tools for analyzing the vast and fast increasing database. After CCIS, need was felt of an application that supports police station operations and the investigation process and that is common across all states and union territories; hence the Common Integrated Police Application (CIPA) was launched in 2005. The core focus of the CIPA application is the automation of police station operations. Its core functionality includes Registration Module, Investigation Module and Prosecution Module. CIPA brought the ability to enter registration (FIR) details into the system and the ability to create and manage police station registers on the system. So far about 2,760 police stations (out of a total of 14,000 þ ) have been covered under the scheme. CIPA project provide basis for the evolution of CCIS. Registration of an FIR with the details of its type, date, mode of information, duty officer, etc. is saved in the CCIS database and can be used for further investigation. The major drawback of CIPA is that it is a stand-alone system which could not provide the enhanced outcomes in the areas of crime investigation and criminals detection that are necessary. Therefore, Ministry of Home Affairs has decided to launch the Crime and Criminal Tracking Network System (CCTNS) program against CIPA. The third evolution is of CCTNS approved by the Cabinet Committee on Economic Affairs on June 19, 2009 with an outlay of Rupees 2,000 crore. This is one of mission

JEIM 27,5

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

518

mode project under the National E-governance Plan . CCTNS aims at creating a comprehensive and integrated system for enhancing the efficiency and effectiveness of policing through adopting of principle of E-governance and creation of a nation-wide networking infrastructure for evolution of IT-enabled-state-of-the-art tracking system around “Investigation of crime and detection of criminals.” The various sub-modules of CCTNS includes registration module, investigation module, prosecution module, basis search module, advanced search module, basis reporting module and MIS module, etc. CCTNS project include all 14,000 police stations in the country and all 6,000 senior police officers. The pilot program under CCTNS has been launched by Home Minister of India on January 4, 2013. CCTNS has also been launched in several Indian states such as Jharkhand and Kerela. The latest evolution is of National Intelligence Grid (NATGRID). This is created in response to new realities of hi-tech-based crime scenarios. Over a period of time, nature of crime has changed which proved earlier systems ineffective in the wake mounting challenge of dealing with terrorisms and naxal movements that has plagued the basic fabric of society. Technology-assisted crimes are believed to have cost the world economy more than $2 trillion last year, far in excess of the Indian GDP of $1.6 trillion (www.mid-day.com/opinion/2011/nov/151111-opinion-Study-policing-for-the-future.htm accessed on 19/12/2011). The 2001 Indian Parliament attack and 2008 Mumbai attacks (also referred to as 26/11 with more than ten coordinated shooting and bombing attacks across Mumbai, India’s largest city) by Islamist attackers were high-profile attack with skilfull use of satellite and wireless communications. Terrorists groups put homework close to perfection. An Inquiry Commission report tabled before the Maharashtra legislative assembly, said the “war-like” attack was beyond the capacity of any police force. Soon after, Prime Minister Manmohan Singh announced the formation of a National Investigation Agency , a federal anti-terrorist intelligence and investigation agency, like the FBI, to coordinate action against terrorism apart from strengthening anti-terror laws with Unlawful Activities (Prevention) Act (UAPA, 2008). Home Minister P. Chidambaram, subsequently mooted a database project called NATGRID to strengthen India’s intelligence infrastructure, which was approved by the Cabinet Committee on Security in May 2011. NATGRID is designed to consolidate and make searchable data gathered by existing security and law enforcement agencies in order to prevent terrorist activity within the country. NATGRID, in its first phase, is aimed to network 21 sets of data sources (in the government and private sector such as banks, insurance companies, stock exchanges, airlines, railways, telecom service providers, chemical vendors, etc.) to provide quick and secure access to information required by ten intelligence and law enforcement agencies as part of the counter terror-related investigative processes. They are the Research and Analysis Wing (RAW), Intelligence Bureau (IB), Central Bureau of Investigation (CBI), Immigration Department, Financial Intelligence Unit (FIU), Central Board of Direct Taxes (CBDT), Directorate of Revenue Intelligence (DRI), Enforcement Directorate, Narcotics Control Bureau (NCB), Central Board of Excise and Customs (CBEC) and the Directorate General of Central Excise Intelligence (DGCEI). These agencies to get bolted-down computer terminals for accessing information from NATGRID while NATGRID to function as a central facilitation centre, to “data sources” such as banks and airlines, telecom companies, hospitals, etc. Agenda of this paper is going to augment the objectives of NATGRID as it proposes an IPS with all the necessary features of an IDSS framework. The interesting part in

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

this is the length and breadth of subject and database being covered in this system. As per NCRB, few crime statistics are given in Tables I and II. These statistics speaks volumes of the complexity and challenging circumstances in which Indian police is working and has to dealing with these crimes and justice system. In view of this it would be a modest claim to emphasize the need of IPSs, details of which will be presented in the following sections. 4. Proposed architecture of IPS The proposed intelligent police information system (IPS) (Chandra et al., 2007; Gupta et al., 2007, 2008) is an outcome of the research project and motivation toward IPS has come after series of interviews and discussion with law enforcement professional. An IPS is required to catch criminals and to remain ahead in the eternal race between criminals and law enforcement. The architecture of IPS is shown in Figure 5. The proposed IPS is designed to meet the current requirement and provide information at all levels of decision making by introducing a multi-level structure of user interface and crime analysis model as shown in Figure 5. It has capability to assist police in the following tasks: .

detecting of crime locations and carrying out crime hot spot analysis;

.

providing information to formulate strategies for crime prevention and reduction; and

.

reducing the further occurrences of similar incidences by analyzing crime patterns.

A framework of intelligent decision support system 519

It shows the architecture of the proposed IPS, which consists of various components. The important components of IPS are Adaptive Query Interface (AQI), Knowledge Acquisition System (KAS), Database Management System (DBMS), Model-Base Management System (MBMS) and visualization. IPS also consists of three storage bases: crime database, model base and knowledge base. Storage bases are uniquely designed to provide requisite data, technique and information at the KAS. The characteristics, role and functionalities of the components and storage bases are given as follows (Figures 6-8). Crime type

No. of reported crimes (2010) % Change from previous year

Indian Penal Code (IPC) Crimes Special and Local Laws (SLL) Crimes Total Cognizables Crimes

Crime type Dacoity Burglary/house breaking Murder Rape Kidnapping and abduction Robbery Riots

2,224,831 4,525,917 6,750,748

No. of reported crimes 1953 2010 147,379 5,579 9,802 2,487 5,261 8,407 20,529

90,179 4,358 33,335 22,172 38,440 23,393 67,571

3.4 2 1.13

Table I. Crime statistics of cognizable crimes

% Change from 1953 to 2010 39 22 240 792 631 178 229

Table II. Comparison of various crime types from 1953 to 2010

JEIM 27,5

User/Decision-Maker DGP/ADGP

IG/DIG District Level (SSP/ASP/DSP) Police Station Level (Inspector/SI)

520

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

Crime Analysis Model Administrative

Strategic

Tactical

Investigative

Web Base Management System Browser 1





Browser n

Web Server

Adaptive Query Interface

Database Management System

• Data storing & retrieval • Addition, deletion & updating of data

Visualization

Model Base Management System

Knowledge Acquisition System

• Model storing & extraction • Model customization as per requirement

(Details in Figure 6)

Model Base

Figure 5. Architecture of intelligent police system (IPS)

Crime Database

Knowledge Base

(Details in Figure 7)

(Details in Figure 8)



• Feature Selection • Clustering, SSC • Classification MTS, DEA, ARM

Knowledge Acquisition System Data Pre-Processing Pattern Evaluation

Crime Data Mining Engine

Figure 6. Knowledge acquisition system

Knowledge

Model Selection Knowledge Base

Various components of intelligent police system (IPS): .

User/Decision-Maker: it is a multi-level user system where users have different role to play at different levels in hierarchy. For example, the rank of Director General of Police (DGP) or Additional Director General of Police (ADGP) with the

A framework of intelligent decision support system

Crime Data Warehouse Extract, Transform, & Load (ETL)

521

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

Crime Database

Entity Based DB

Missing Person DB

Crime Stat. DB

Figure 7. Indian crime database Raw Data Sources

Domain Expert Knowledge

Historical Events

Knowledge Base

Police Doctrines

Earlier Results and Patterns

responsibilities of administration of the state police force, are interested in knowing crime trends and hot spots in the whole state, whereas district level police is interested in the performance police station within its area. .

Crime Analysis Model: the crime analysis model provides accessibility to various functionalities of IPS as per the Police hierarchy since roles differ at each level of the police force. Four types of crime analysis models namely administrative, strategic, tactical and investigative have been proposed to meet the requirements of all four levels of decision makers in Indian Police system. The decision makers at the higher levels (DGP/ADGP) can access the lower levels of the crime analysis model but users at the lower levels (Police Station) cannot access the higher levels (strategic models). The user/decision maker is connected to a webbased management system.

.

Web base management system: this facilitates information accessibility and exchange throughout a state. WBMS is needed to coordinate and communicate from anywhere in the state since state is sparsely populated over a geographical region and data as well as information are available at various location. WBMS

Figure 8. Knowledge base

JEIM 27,5

allows the user to extract the relevant information from a distant location using uniquely designed Adaptive Query Interfaces (AQI). .

Adaptive Query Interface: Adaptive Query Interface (AQI) provides a userfriendly adaptive interface between the user and the knowledge acquisition system. AQI is designed for supporting multi-level user interface. For example, in crime hotspot analysis, area of interest changes with respect to crime analysis model. It means that district level police officer can find crime hotspot at police station level whereas stat level police officer can also get crime hotspots at district level. The objective of AQI is to shift functionality as much as possible from the user to the system. It provides the user a fast way to carry out the process of identification of crime hot spots and crime zones. The Knowledge Acquisition System (KAS) identifies the most suitable data mining tasks and techniques by adaptive query interface.

.

Knowledge Acquisition System: Knowledge Acquisition System (KAS) works as brain of IPS as several data mining techniques and algorithms for knowledge acquisition have been considered in KAS. It has links to every subsystem of IPS. Figure 6 shows the steps involved into KAS. Data are acquired from crime database using Database Management System (DBMS) and is pre-processed in the system. KAS applies the most appropriate model and algorithm of data mining from the model-base using Model Based Management System (MBMS) based on the query obtained from the adaptive query interface. Clustering, Classification and Association Rule Mining are selected based on the type of crime analysis. The algorithm is also selected based on nature of the data, e.g. categorical, continuous, entity, etc. The selected algorithm is applied to find out the desired results from the pre-processed data in the crime data-mining engine. The results are further evaluated to generate knowledge in terms of interesting crime patterns. The knowledge base also provides inputs to the crime data-mining engine and the pattern evaluation process. The knowledge generated by KAS is passed to the visualization module and also stored in the knowledge base for future usage. If desired results are already in the knowledge base, computation can be skipped.

.

Crime Database: it deals with an increasing number of crimes – approximately 22 lakh, Indian Penal Code (IPC) crimes and 45 lakh local and Special Law crimes per year. These crime data are collected from various sources such as passport department, transport and crime records. These are integrated into the crime database. NCRB stores them in multiple databases depending upon their applicability. CCIS (as entity-based database), Crime in India (crime statistics database) and Talash (missing person database) are three important databases maintained by NCRB (Figure 7). These databases are sources from raw data sources such as first information report (FIR). In order to give multi-dimentional view of the data to the decision maker, a data warehouse has been designed based on these three available data sources. Crime database indirectly connects to knowledge acquisition system via specifically designed database management system (DBMS) so that all three levels of data can be used appropriately.

.

Database Management System: this controls the organization, storage, management and retrieval of data in the crime database. The type of query decides the selection of a particular database in the IPS. The traditional way of

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

522

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

.

.

.

executing queries is extracting records and aggregating them for every execution. This is a highly time consuming process because the database needs to be rescanned every time and it requires a specialized user to extract the information from the database. DBMS for crime database eliminates rescanning of the database for every new query as well as the need for skilled users. It provides a fast way to extract relevant data from the crime database. The joining of many relational tables is also required for extracting crime data against some specific query since not all information required by the query is stored in a single table. The proposed DBMS for crime database extracts the records from these tables and aggregates them for further online querying. Model Base Management System: Model Base Management System (MBMS) stores and extracts a large amount of models used for crime analysis. It plays an important role since it extracts a model or a combination of models as per requirement. It is difficult to manage various kind of models used for IPS since each model is applied in a specific problem domain and also has specific data requirements. For example, clustering is used for crime hot spot analysis with data requirements of location vs crime densities. On the other hand, DEA with clustering is used to rank police administration units on the basis of data pertaining to crime prevention measures. MBMS provides the appropriate model or combination of models with specific data requirements to KAS. Model Base: model base stores data mining and performance measurement models such as Clustering, Semi-Supervised Clustering, Classification, Association Rule Mining and Data Envelopment Analysis (DEA). Each model has a number of algorithms based on the type of analysis to be carried out. Clustering algorithms ( Jain et al., 1999) like K-Means, Hierarchical, etc. are used for identifying crime zones and crime hot spots. Decision tree’s algorithms like Classification and Regression Trees (CART) (Breiman et al., 1984) and multilayer perceptron model (MLP) (Haykin, 2008) of Artificial Neural Network etc. have been stored as Classification models. Association rule mining algorithms (Agrawal et al., 1993), for finding frequent item sets whereas, Fast Algorithm for mining association rules have been stored in the model base for investigation of criminal cases, criminal profiling, apprehension of criminals and establishing linkages between crime entities and criminals. Malmquist DEA model (Charnes et al., 1978; Tone and Tsutsui, 2010) as performance measurement technique has been applied in combination with clustering technique to rank police administration units on the basis of their effective enforcement of crime prevention measures. Knowledge Base: this consists of four integral components, i.e. domain expert knowledge, police doctrines, historical events and results and patterns obtained earlier as shown in Figure 8. The police analyst and their knowledge can be stored in the forms of rules into the knowledge base. The police manuals and doctrines as decision rules also contribute to the knowledge base. Historical events and occurrences as a component of knowledge base are used in investigative and tactical crime analysis model for comparing and correlating forthcoming information in the light of historical events. Whenever the user fires a similar query, earlier results and patterns component in the knowledge base provides the desired results instantly instead of going through the whole process of obtaining results again. The objective of knowledge base is to enhance usability of the information and pattern retrieval process.

A framework of intelligent decision support system 523

JEIM 27,5

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

524

.

Visualization: by using visualization, IPS summarizes the results, patterns and information and also highlights trends and phenomena through various kinds of graphs, tables and map representations. Crime trends plots are helpful in predicting the future occurrences of crime at various locations. Tabular display and graphs are helpful in analyzing crime densities and also comparing with different areas of interest. Maps display enhances the understanding of crime hot spots and identification of crime zones. Police patrolling in sensitive areas can be increased as per the maps and graphs. The visualization system plays a significant role in effective enforcement of law and order. The user can enhance the understanding of results using any kind of Geographical Information System (GIS).

5. Current status of IPS prototype on Indian crime records As a proof-of-concept and to demonstrate the utility of IPS as an IDSS for the Indian police, we have implemented a prototype of IPS on Indian crime records at NCRB. The IPS prototype incorporates the crime analysis model, an AQI, KAS, DBMS for crime database, MBMS along with a model base which stores several models required for crime data mining and performance measurement and visualization modules to provide information to the end user. The remaining systems such as Web-Base Management System (WBMS) and geographical information system (GIS) will be implemented in the future course of action. The details of each component and its significance are given in the subsequent section of the paper. 5.1 Implementation of crime analysis model The crime analysis model has been implemented to cater to the need of policing at different levels of police hierarchy since role differs at each level of the police force. Users can access various crime analysis models as per their rank. The top level official can access the crime analysis model used by lower rank officers: the DGP and ADGP have the accessibility to all four type of crime analysis models, i.e. administrative, strategic, tactical and investigative as shown in Figure 9, whereas the Inspector and Sub-Inspector can access only the investigative type crime analysis model.

Figure 9. Crime analysis models accessibility in IPS as per rank

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

Different kinds of interfaces have been designed to cater to the need of all type of crime analysis model. A brief description of crime analysis models is given below. The investigative crime analysis model has been developed to obtain patterns that will assist in linking together and solving current serial criminal activity. This type of analysis focusses primarily on qualitative data comprising serial crimes such as murder and rape. Criminal profiling is the main task of this analysis model. The police personnel at the police station and group of police stations level are often interested in knowing such patterns for investigation and criminal profiling. In order to identify crime patterns and crime trends for recent criminal and potential criminal activity at district or sub-district level, tactical crime analysis has been developed. The features of tactical crime analysis are linking cases together and identifying the notable characteristics of patterns and trends. It is highly important to know crime trends for those districts which have been designated as crime hot spots. In order to study the crime trends at various crime zones, a separate GUI has been developed in IPS. The user can obtain the crime trends plot of a particular district under certain crime type for the specified period using the designed GUI for crime trends as shown in the Figure 10. Figure 10 presents the crime trends of district 119 with respect to crime type murder (homicide) from year 2001-2006. Figure 10 shows decreasing crime rate for district 119 from 2001 to 2006. District 119 is still a crime hot spot even though it is showing decreasing crime trends since the average crime density of district 119 is greater than the average density of districts in the other crime zones. The user can obtain similar crime trends under the tactical crime analysis model of the proposed IPS. The strategic crime analysis model is used to determine long-term patterns to formulate strategies such as deployment of police force, patrolling, etc. For example,

A framework of intelligent decision support system 525

Figure 10. Crime trends plot for district “119” under murder (Homicide)

JEIM 27,5

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

526

it is found that there are a greater number of thefts and other crimes during daytime in a locality of one of the four metropolitan cities in India. Further analysis that has been carried out over past data has led to the conclusion that most couples are working in that locality, hence no one is in the house at the time of theft. Therefore, this kind of analysis helps in forming the strategy of deploying more police force during the day so that occurrences of such incidences of crime can be prevented. In IPS, the strategic analysis model can be accessed by police officers at the district level or at the level of a group of districts. They can analyze the crime density of the specified crime type for a desired period. Crime zones and crime hot spots can be identified using proposed AQI as described in detail in the subsequent section. The results of this query interface provide insight for formulating the strategy of police deployment, patrolling, etc. Administrative crime analysis is primarily designed to cater to the needs of higher levels of police hierarchy. State level police officers such as DGP and ADGP can generate summaries and reports of state crime, which can be used to administer the police force and also to provide information to the general public, the media, etc. They can also obtain the information of crime hot spots and crime zones within a state using the proposed AQI in IPS. This information is helpful for administrating the state police force and enforcing law and order at state level. The AQI designed for fulfilling the needs of the top three levels of the crime analysis model is discussed in details in the following section. 5.2 Implementation of AQI An AQI is designed and implemented to provide easy and user-friendly access to the system so that various patterns and information from the data can be obtained. The online AQI has been especially proposed for mining of crime data. The top three levels of the crime analysis model, i.e. administrative, strategic and tactical analysis have the common problem of identifying crime hot spots and crime zones of a particular region of interest for certain crime types for specified periods. The proposed interface is designed for finding solutions to the above-mentioned problems. The proposed interface can further be used to analyze crime trends with respect to individual crime type of crime hot spots so that prediction of crime trends can be made. Clustering algorithms have been applied for identification of crime hot spots and crime zones. The effectiveness of the proposed AQI has been illustrated on the database of Indian crime database at NCRB using a developed software tool. The proposed interface will also be helpful in analyzing crime and controlling it. The interface can play an important role in tackling a wider variety of similar problems. The description of the proposed crime analysis interface for the top three levels of crime analysis models is given below. Crime Analysis Wizard like AQI will identify crime hot spots and crime zones of a particular region for certain crime types for a specific period. This interface will provide a tool for making an online query and based on the query, crime hot spots will be identified. The user can select a particular state/UT from the list. The list contains only those states/UTs for which data extraction has taken place. A user can select any year as well as multiple years with the help of Year Selection Input Dialog Box, which will appear after “select a year.” The module also has the facility to select crime type whether from the 105 listed crime types by selecting Selection Type as Crime Type or from one of the major crime heads, i.e. crime against body, crime against women, crime against property and kidnapping and abduction by selecting Selection Type as Crime Head. Crime Head as selection type helps the layman user to provide inputs to KAS for analyzing crime hot spots and crime zones of general type of crimes without going into specific details.

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

The results of queries could be crime hot spots, high crime zones, moderate crime zone and low crime zones based on the average density of these crimes. Cluster centres as average density of various crime zones are also given in the results of the query interface. Figure 11 shows the average crime density of crime hot spot districts of the state of Karnataka under seven major crimes, i.e. abduction, dacoity, hurt, kidnapping, murder, rape and robbery for year 2006. Similar kind of results can easily be obtained by various levels of police administrators. The results obtained using the proposed AQI will be helpful in identifying crime hot spots and predicting crime trends in crime hot spots which will ultimately help in controlling crime. The user can view the results of the state and any of its districts by selecting a particular district from the options provided in the result panel. The user can generate a summary and report against any query for further analysis and framing policies. The user can further analyze the result using Data Information and Comparison Wizard, which is described in the visualization section of the paper. The results obtained represent not only crime hot spots but also various crime zones in the area. Identification of crime hot spots can be helpful in reducing crime in the area by providing special attention to such police stations. This information can be used for bringing out policies for reducing crime in the area.

A framework of intelligent decision support system 527

Figure 11. Adaptive query interface for crime hot spots analysis

JEIM 27,5

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

528

The analysis of results is also helpful in predicting future crime hot spots in addition to the existing ones in the form of identifying police stations in high crime zones, which are in proximity to police stations in very high crime zones, i.e. which are going to become crime hot spots. Information about future crime hot spots is used to reduce the crime in the area. These crime trends are also helpful in measuring the performance of crime preventive measures based on whether a particular type of crime has increased or decreased. The proposed interface can also be utilized for mining of crime data at the police station level. This tool not only mines crime data in a user interactive manner but also enhances the compatibility of analyzing crime data. The user can also obtain crime trends of crime hot spot districts for a certain crime type. By analyzing crime trends, the user can predict crime density for a location and take preventive measures for future occurrences of crime. For the given query in Figure 11, districts “103,” “105,” “107,” “115,” “119,” “125” and “132” have been identified as crime hot spots. Figure 12 shows the crime trends of crime hot spots, e.g. districts “103,” district “107” and “132” with respect to individual crime types. Figure 12 shows an increase in dacoity in district “103” for the period 2000-2004. This information can be used to prompt the police administrating units that necessary actions are required to reduce the number of dacoities in district “103.” It is seen from other plots of Figure 12 that there are decreasing trends for different crime types. The districts are still designated as crime hot spot since average crime density is greater than most other districts. Crime Trends Plot

Crime Trends Plot

110

110 100

90

90 Crime Density

Crime Density

District “103” under Dacoity

100

80 70 60

80 70 60 50

50 40 2000

District “107” under Murder

40 2001

2002

2003

30 1998 1999 2000 2001 2002 2003 2004 2005

2004

Year

Year

Crime Trends Plot

Crime Trends Plot

14

16

District “132” under Rape

District “132” under Suicide

12

14 Crime Density

Crime Density

10 8 6 4 2

Figure 12. Crime trends plots

0 2001

12 10 8 6 4

2002

2003

2004 Year

2005

2006

2 2002

2003

2004 Year

2005

2006

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

It is to be noted that data needs to be extracted at least once before for carrying out crime analysis using AQI. Data extraction need not be run every time for carrying out crime analysis on the same data for the same period; this is required to be done only the first time to extract records. For example, if the user has already extracted the records of Karnataka for the period 2000-2006 then he does not need to run Data Extraction Wizard for the same state and same period. But if he would like to run a query for Punjab or any other state, he has to run the Data Extraction Wizard for extracting records of the desired state. The details of crime database used in IPS and data extraction wizard are given in the next section.

A framework of intelligent decision support system 529

5.3 Implementation of data extraction module for Indian crime database IPS has been implemented on various crime databases as well as the data warehouse maintained by NCRB. Crime database used in IPS contains the necessary data for data analysis and mining activities. IPS has been used on crime data of Karnataka at the initial stage and later extended for carrying out analysis on all Indian crime data. Karnataka has more than 20 lakh crime records till 2006. The proposed IPS has performed remarkably well in handling such a database and carrying out crime analysis for Karnataka. A separate Data Extraction Utility has been designed to facilitate the extraction of data and updating the already extracted records for carrying out crime analysis. The description of Data Extraction Wizard has been given below. Figure 13 shows Data Extraction Wizard that establishes the connection between the database of Indian crime database and IPS and also extracts records state-wise for carrying out further crime analysis. An ODBC data source is required before running this module to create a link from the database of Indian crime database. To establish the database connection some inputs such as database name, username and password are required from the user. The extraction of records is also carried out through this wizard. The user is provided many options to facilitate him/her in extracting records since extraction of data from a huge crime database requires a lot of time. The user can select any State/UT or all states from the popup menu provided in the Extract Records panel of the wizard. After state selection, Year Selection Input Dialog

Figure 13. Data extraction wizard

JEIM 27,5

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

530

Box will appear which displays the years (in descending order) for which data are present; an error dialogue box appears if data are not present. A user can choose starting year from which he/she is interested in carrying out analysis. For example, if Year Selection Input Dialog Box displays years from 2006 to 2000 and the user selects 2004 as starting year, then data from 2004 to 2006 will be extracted. Similarly, if Year Selection Input Dialog Box display years from 2003 to 2000 and the user select 2001 as starting year, then data from 2001 to 2003 will be extracted. After year selection, RUN button is to be pressed to extract data records of the selected State/UT in the specified period. UPDATE button is pressed if new data has come for subsequent year. It will update already extracted data for the same State/UT; last three years data would be extracted if no extracted data were present. Take, for example, that the user has already extracted the records of Karnataka for the period 2000-2006 and new data for year 2007 has arrived. If the user wants 2007 data for crime analysis, he can only select year 2007 as starting year and press UPDATE and it will update already extracted data. UPDATE option will save a lot of time for future records since it will run a query on the latest data and update the already extracted data files. Once data records are extracted then a model is selected from the model base of IPS to obtain desired information. The details of model base designed in IPS and accessibilityrelated issues have been described in next section. 5.4 Model base for IPS The model base has been specially designed to store the large number of models used for crime analysis. The details of each model and its algorithms used in the model base along with application areas and data requirements are described below. The feature selection technique has been used to extract significant features from the huge volume of crime records which increase at the rate of approximately 22 lakh Indian Penal Code crimes and 45 lakh local and Special Law crimes per year. The feature selection algorithm not only selects significant features but also provides feature weights since all types of crimes do not have equal weights – for example, murder will have more weights over kidnapping and hurt. ARM is applied for investigation of criminal cases, criminal profiling, apprehension of criminals and establishing linkages between crime entities and criminals. It is also used for finding the co-occurrences of various crimes together so that controlling one such crime will reduce the occurrence of another related crime. Quantitative ARM has also been used to address the problems of numerical data and incoming crime records, respectively. Classification algorithms, namely decision tree, Artificial Neural Network (ANN) approach, Bayesian approach, etc. have also been stored in the model base for catering to various applications in crime data mining. Crime zones can be treated as classes for supervised learning. The various classification models have been applied as per the application areas. Clustering algorithms are used for identifying crime zones and crime hot spots. Parametric Minkowski distance measure have also been applied to introduce the weightage scheme in the clustering process since all types of crime do not have equal weights: murder, for example, will have more weights over kidnapping and hurt. These weights can either be obtained by the user’s input or by applying standard algorithms of weight generation and selection. SSC approaches have also been used in model base for IPS. SSC uses a small number of labeled objects to improve unsupervised clustering algorithms. SSC algorithm is used

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

to cluster crime locations and to find crime hot spots in India, as the density of crime incidents will be continuous over an area, higher in some parts and lower in others. For this application, NCRB experts have provided the information about crime zones of a few locations based on their domain knowledge. Performance analysis is another crucial task for effective policing. DEA as a performance measurement technique has been applied in combination with clustering technique to rank PAUs on the basis of their effective enforcement of crime prevention measures. MTS clustering in combination of dynamic DEA has been carried out on periodic crime data. MTS clustering forms homogenous groups of PAUs such as states, districts and police stations based on similar crime trends. Dynamic DEA is further applied to rank PAUs on the basis of their effective enforcement of crime prevention measures. 5.5 Performance analysis using Malmquist DEA model We have identified only the relevant input and output measures from crime prevention measures before applying DEA. The selected input measures for analysis are civil and armed police force strengths and total police expenditure because a PAU has high efficiency if it is utilizing its resources well both personnel and financial resources. Total police expenditure consists of all the money invested for implementing all crime prevention measures. Output measures are selected by considering the fact that DEA model tries to maximize the output measures to make DMUs as efficient. Therefore, apprehension of criminals, i.e. number of person arrested is considered as the first output measures for analysis. The second output measure is the inverse of crime rate since efficiency is inversely proportional to crime rate. Here, crime rate is obtained by dividing crime density of the state/UT with total population of that state/UT. It is observed that variance is quite high for every input and output measures which indicates that the data is not homogenous in nature. In such a case, DEA does not produce desirable results. Therefore, MTS clustering is required to cluster the data into homogenous groups prior to applying DEA. A comparative evaluation of results with MTS clustering and without MTS clustering is shown here. The results without MTS clustering has been obtained when all 35 states/UTs are considered as one homogenous group for Malmquist DEA model whereas the results with MTS clustering are obtained when the states/UTs of three crime zones (namely low, high and medium) are considered separately as homogenous group for Malmquist DEA model. Figure 14 shows the comparative plots of MPI with and without MTS clustering. It is seen in Figure 14 that MPI without MTS clustering is higher than MPI with MTS clustering for high and moderate crime zone states irrespective of their poor performance. These states are Chhattisgarh, Haryana, Jharkhand, Orissa, W.B., Kerala, M.P., Maharashtra, Rajasthan, T.N. and U.P. We have selected two states namely, U.P. and Jharkhand to demonstrate that the results with MTS clustering are better than the results without MTS clustering. Figure 15 shows the MPI with MTS clustering and without MTS clustering for U.P. state. The performance is inversely proportional to the crime rate. Crime rate of U.P. is shown in Figure 16. From Figures 15 and 16, it is seen that MPI with MTS clustering is 1.63 for the year 2002-2003 whereas MPI without MTS clustering is 1.05 which correlates with the fact that there is a decrease in crime rate in UP during the period 2002-2003 from 85.5/lakh to 54.5/lakh. But for year 2003-2004, MPI with MTS clustering is decreased from 1.63 to 0.87

A framework of intelligent decision support system 531

JEIM 27,5

Comparative MPI

MPI Without MTS Clustering

MPI With MTS Clustering

1.6 1.5

MPI

1.4

532

1.3 1.2 1.1

Figure 14. State wise comparison of MPI with and without MTS clustering

A.P. Ar.P. Assam Bihar Chhattisgarh Goa Gujarat Haryana H.P. J&K Jharkhand Karnataka Kerala M.P. Maharashtra Manipur Meghalaya Mizoram Nagaland Orissa Punjab Rajasthan Sikkim T.N. Tripura U.P. Uttaranchal W.B. A&NI Chandigarh DNH D&D Delhi Lakshadweep Pondicherry State/UTs

Malmquist Productivity Index: U.P. 2.00

MPI with clustering

1.80

MPI without clustering

MPI

1.60 1.40 1.20 1.00 0.80 2002-03

Figure 15. MPI of U.P.

2003-04

2004-05 Time Period

2005-06

2006-07

Crime Rate: U.P. No. of Crime /lakh people

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

1.0

Figure 16. Crime rate of U.P.

90 80 70 60 50 40 30 20 10 0 2002

2003

2004

2005

2006

2007

Year

due to the increase in crime rate from 54.5/lakhs in year 2003-73.2/lakh in year 2004. Therefore, the results with MTS clustering reflect the actual performance measure whereas the results without MTS clustering does not commensurate the outcome. Similarly, the results for Jharkhand have been separately shown to highlight

A framework of intelligent decision support system 533

Malmquist Productivity Index: Jharkhand 2.20

MPI with clustering

2.00

MPI without clustering

MPI

1.80 1.60 1.40 1.20 1.00 0.80 2002-03

2003-04

2004-05 Time Period

2005-06

2006-07

Figure 17. MPI of Jharkhand

Crime Rate: Jharkhand No. of Crimes /lakh people

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

the fact that results with MTS clustering is more efficient as compared to the results without MTS clustering. Figure 17 shows the MPI with MTS clustering and without MTS clustering for Jharkhand state. Crime rate of Jharkhand is shown in Figure 18. For Jharkhand state, MPI with MTS clustering is decreased from 1.22 in year 2003-2004 to 1.01 in year 2004-2005 since there is an increase in crime rate from 110.5/lakh in the year 2004 to 121.8/lakh people in the year 2005. The fact that results with MTS clustering are far superior as compared to the results without MTS clustering is true for all states. The experimental results show the superiority of MTS clustering over the results without MTS clustering. The approach will offer a computer-based environment not only to rank PAUs but also provide an evaluation tool to monitor the implementation of crime prevention measures at various levels of police administration on regular basis. A Model Base Wizard has also been designed to add and delete algorithms of any model in the model base so that improved or hybrid models can be assimilated in future. Only the administrator can access Model Base Wizard since the administrator knows about the usage of various models: laymen need not know about the models for analysis. If laymen had been provided the facility, it could be futile in IPS. Figure 19 shows the utility of adding ROCK (Robust Approach of Clustering using Links) as one of the new clustering algorithm for data mining. The administrator can also provide various characteristics such as type of data handling, time complexity, space complexity and high, moderate and low accuracy.

135 130 125 120 115 110 105 100 2002

2003

2004 2005 Year

2006

2007

Figure 18. Crime rate of Jharkhand

JEIM 27,5

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

534

Figure 19. Model base wizard

This information is helpful in the model selection process in KAS. The selected algorithm is applied to find out the crime zones and crime hot spots out of the pre-processed data in the crime data-mining engine. The knowledge base also provides inputs to the crime data-mining engine. The results generated by KAS are passed to the end user through a visualization module. The details of the visualization subsystem of IPS are described below. 5.6 Implementation of visualization Data Visualization Wizard has been designed to provide visualization of results to the end-user. It displays comparative crime plotting of specified query fired in the crime analysis wizard. It displays the plot for a particular type of crime selected from the Crime Type popup Menu. In the given example, a crime plot of one of the district of Karnataka under murder crime is shown in the Figure 20. It displays crime plots of one of the district of Karnataka for the years 2002-2005 under murder crime. It can be observed that one police station has a greater number of occurrences of murder in comparison with other police stations. The user can also see the pie chart of any district depicting several crime types. Figure 21 shows the pie chart of Bangalore district under seven crimes against body. Figure 21 shows that hurt crime has the largest occurrence as compared to others. This information can be utilized to reduce crime incidence by taking crime preventive measures in those police stations in which crime occurrences are large. Therefore, visualization facility in IPS not only provides desired results to the user but also provides information of crime incidences at each level of police administration. Above descriptions are still too short to do justice in terms of describing entire range of capabilities of IPS. Results shown are from real data provided by NCRB.

A framework of intelligent decision support system

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

535

Figure 20. Data visualization wizard

Figure 21. Crime chart of Bangalore city

Many modules are under use with the funding agency (NCRB). IPS follows a natural path of evolution and captures current realities of crime management. A comparison is made in Table III of IPS with its precursors and contemporary systems over key parameters like type, objective, unit of operation, etc. IPS is found quite contemporary and provides the much needed multi-level user interface, analytical tool, etc. for carrying out police activities more efficiently.

Database Management System

To store and retrieve crime records

District

Not available

Not available

Type

Basic objective

Basic unit of operation

Multi-level user interface

Analytical tool

Table III. Comparative analysis of existing vs proposed police information system

CCIS

Not available

Not available

Police station

To build CCIS automatically and to automation of all functions carried out at primary source of information

Information System

CIPA

Not available

Available

Integrated database management System covering only police department entities To create a comprehensive and integrated system for enhancing the efficiency and effectiveness of police at all levels including at Police station level Police station

CCTNS

Available

Available

Available

Available

Police station

To meet and fulfill the new emerging responsibilities and tasks of the police

To network 21 available databases across government and private agencies to “flag potential terrorist threats”

Database centre of the organization

Decision Support System

IPS

Integrated database management System covering 22 department entities

NATGRID

(continued)

CIPA and IPS are designed to cater to the need of the basic unit of police administration, i.e. the Police Station Crime analysis model is used to meet the requirement of police hierarchy since the role differs at each level of the police force Data mining techniques provide analytical facility in IPS

Till date in CIPA, only General Diary and Registration Modules are operational whereas other important modules, namely Investigation, Prosecution, Information, Reports and Queries and Supervision Module are still to be operational

IPS is based on current data mining, performance measurement techniques and decision support technologies

Remarks

536

Description

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

JEIM 27,5

CCIS

Not available

Not available

Description

Utility of crime hot spot analysis

Utility of performance analysis

Not available

Not available

CIPA

Available

Available

CCTNS

Available

Available

NATGRID

Available

Available

IPS

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

Clustering and semisupervised clustering are used for crime hot spot analysis in IPS Clustering technique and data envelopment analysis (DEA) are applied to monitor the performance on a regular basis in IPS

Remarks

A framework of intelligent decision support system 537

Table III.

JEIM 27,5

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

538

6. Conclusion This research was supported by India’s NCRB with specific mandate to develop the architecture of IPS having capability of analytical tool in combination with key utilities like crime hot spot analysis and performance analysis, for which advance data mining tools and DEA have been suitably tested and integrated. This research is a timely response to meet the emerging need of building analytical power into the existing police systems. As is known, crime and control system are required to pool and feed data from myriad of sources in order to meet the requirements of justice system. The main contribution of the paper is the framework for police DSS with specific reference to Indian police force, which is quite large in size and deals with high number of crime under strict police hierarchy. This framework has several intelligent features which are illustrated with crime data in Indian settings characterized by large volume and high rate of crimes that are incomparable any police force elsewhere in the world. It is named IPS, architecture of which is based on the current decision support and data mining techniques in order to carrying out police activities efficiently. Effectiveness of the above system has been tested through a prototype at NCRB for crime hot spot analysis and performance measurement of PAUs. The present focus of the study was to handle large voluminous crime data and analyzing it to generate meaningful crime patterns. It holds promise of an effective tool to law enforcement agencies for crime detection and crime prevention. Some important additional features include WBMS, extended visualization module using GIS (for enhancing the understanding of results and patterns), mobile ad hoc network technology, etc. These would enhance its usability at remote locations. The IPS can be scaled up in capability for providing real-time analysis which requires dedicated infrastructure and resources for monitoring real-time events. The analytic capability of the system can also be enhanced on real-time basis over additional facilities such as GIS, web-enabled tools and provision of specialized task force by law enforcement agencies. The IPS can also cater to the needs of future requirement of law enforcement agencies by model enhancement utility. IPS can easily be applied to any police system in the world and can equally be useful for any law enforcement agencies for carrying out homeland security effectively. For the researchers, it would be useful to attempt similar framework in other law enforcement agencies like para military forces, coast guard, etc. For example, coast guard collects massive amount of electronic intelligence data, which not only needs to be analyzed on a regular basis but also requires to generate crime patterns in order to enforce coastal law effectively. This paper is useful for any future attempt in these directions. References Agrawal, R., Imielinski, T. and Swami, A.N. (1993), “Mining association rules between sets of items in large databases”, Proceedings of the ACM SIGMOD International Conference on Management of Data, Washington, DC, pp. 207-216. Bella, P., Deana, G. and Gottschalkb, P. (2010), “Information management in law enforcement: the case of police intelligence strategy implementation”, International Journal of Information Management, Vol. 30 No. 4, pp. 343-349. Bhargava, H.K., Power, D.J. and Sun, D. (2007), “Progress in web-based decision support technologies”, Decision Support Systems, Vol. 43 No. 4, pp. 1083-1095. Bilenko, M., Basu, S. and Mooney, R. (2004), “Integrating constraints and metric learning in semi-supervised clustering”, Proceedings of International Conference on Machine Learning (ICML), pp. 81-88.

Boba, R. (2001), Introductory Guide to Crime Analysis and Mapping, US Department of Justice, Washington, DC.

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

Breiman, L., Friedman, J., Olsen, H.R.A. and Stone, C.J. (1984), Classification and Regression Trees, Chapman and Hall, New York, NY. Chandra, B., Gupta, M. and Gupta, M.P. (2007), “Adaptive query interface for mining crime data”, Lecture Notes in Computer Science (LNCS), Vol. 4777, pp. 285-296, available at: http://link.springer.com/chapter/10.1007%2F978-3-540-75512-8_20 Chandra, B., Gupta, M. and Gupta, M.P. (2008), “A multivariate time series clustering approach for crime trends prediction”, IEEE SMC Conference, Singapore, pp. 892-896. Charnes, A., Cooper, W.W. and Rhoades, E. (1978), “Measuring the efficiency of decision making units”, European Journal of Operational Research, Vol. 2 No. 6, pp. 429-444. Chen, H., Schroeder, J., Hauck, R.V., Ridgeway, L., Atabakhsh, H., Gupta, H., Boarman, C., Rasmussen, K. and Clements, A.W. (2003), “COPLINK connect: information and knowledge management for law enforcement”, Decision Support Systems, Vol. 34 No. 3, pp. 271-285. Cheung, W., Leung, L.C. and Tam, P.C.F. (2005), “An intelligent decision support system for service network planning”, Decision Support Systems, Vol. 39 No. 3, pp. 415-428. Chung, W., Chen, H., Chang, W. and Chou, S. (2006), “Fighting cybercrime: a review and the Taiwan experience”, Decision Support Systems, Vol. 41 No. 3, pp. 669-682. Dy, J. and Brodley, C. (2004), “Feature selection for unsupervised learning”, Journal of Machine Learning Research, Vol. 5, January, pp. 845-889. Gupta, M., Chandra, B. and Gupta, M.P. (2007), “Crime data mining for Indian police information system”, Proceedings of 5th International Conference on E-Governance (ICEG), Hyderabad, pp. 388-397. Gupta, M., Chandra, B. and Gupta, M.P. (2008), “Ranking police administration units on the basis of crime prevention measures using data envelopment analysis and clustering”, Proceedings of 6th International Conference on E-Governance (ICEG), IIT Delhi, pp. 40-53. Gupta, M.P. and Mitra, R.K. (2005), “Assessing internal efficiency, employees and public satisfaction in police E-governance”, 3rd International Conference on E-Governance, pp. 77-84. Han, J. and Kamber, M. (2006), Data Mining: Concepts and Techniques, 2nd ed., Morgan Kaufmann, San Francisco, CA. Haykin, S. (2008), Neural Networks: A Comprehensive Foundation, 2nd ed., Macmillan, New York, NY. Jain, A.K., Murty, M.N. and Flynn, P.J. (1999), “Data clustering: a review”, ACM Computing Surveys, Vol. 31 No. 3, pp. 264-323. Janssen, M. and Cresswell, A.M. (2005), “An enterprise application integration methodology for E-government”, Journal of Enterprise Information Management, Vol. 18 No. 5, pp. 531-547. Krishnamorthy, S. (2003), Preparing the Indian Police for 21st Century, Puliani and Puliani, Bangalore. Levine, N. (2002), CrimeStat 2.0: A Spatial Statistics Program for the Analysis of Crime Incident Locations, US Department of Justice, Washington, DC. Mitra, R.K. and Gupta, M.P. (2008), “A contextual perspective of performance assessment in E-government: a study of Indian police administration”, Government Information Qtly, Vol. 25 No. 2, pp. 278-302. Oatleya, C.G. and Brain, W.E. (2003), “Crimes analysis software: ‘pins in maps’, clustering and bayes net prediction”, Expert Systems with Applications, Vol. 25 No. 4, pp. 569-588. Peng, Y., Zhang, Y., Tang, Y., Power, D.J., Sharded, R. and Carlssone, C. (2011), “An incident information management framework based on data integration, data mining and multicriteria decision making”, Decision Support Systems, Vol. 51 No. 2, pp. 316-327.

A framework of intelligent decision support system 539

JEIM 27,5

Downloaded by Indian Institute of Technology Delhi At 10:23 22 September 2014 (PT)

540

Shim, J.P., Warkentin, M.J., Courtney, F., Power, D.J., Shardad, R. and Carlssone, C. (2002), “Past, present and future of decision support technology”, Decision Support Systems, Vol. 33 No. 2, pp. 111-126. Siskos, Y. and Spyridakos, A. (1999), “Intelligent multi-criteria decision support: overview and perspectives”, European Journal of Operational Research, Vol. 113 No. 2, pp. 236-246. Teng, J.T.C., Mirani, R. and Sinha, A. (1988), “A unified architecture for intelligent DSS”, Proceedings of 21st Annual Hawaii International Conference on Decision Support and Knowledge Based Systems Track, pp. 286-294. Tone, K. and Tsutsui, M. (2010), “Dynamic DEA: a slacks-based measure approach”, Omega, Vol. 38 Nos 3/4, pp. 145-156. Turban, E. (1995), Decision Support and Expert Systems: Management Support Systems, Prentice Hall, Englewood Cliffs, NJ. Weerakkodya, V., El-Haddadeha, R., Sabolb, T., Ghoneima, A. and Dzupkab, P. (2012), “E-government implementation strategies in developed and transition economies: a comparative study”, International Journal of Information Management, Vol. 32 No. 1, pp. 66-74. About the authors Dr Manish Gupta is currently working as the Vice President-Analytics at Info Edge (India) Ltd (Naukri.Com), Noida and responsible to research and develop state of art analytics solutions. He holds a PhD in Data Mining from the Department of Mathematics, Indian Institute of Technology Delhi with over 15 research/technical publications in leading international journals and conferences with one US Patents (filed). He has previously worked as an Assistant Vice President, Citigroup, Bangalore; Principal Analytics Consultant (Head R&D), Innovation Labs@24/7 Customer, Bangalore and Scientist in DRDO, Delhi. He is the recipient of Lab Scientist of the Year Award-2008 and Technology Group Award-2008 and 2009. His research interests include Intelligent Decision Support System, Data Mining, Clustering and Artificial Neural Network. B. Chandra is a Professor in the Department of Mathematics, Indian Institute of Technology Delhi. She has been a visiting Professor for a year at the Graduate School of Business, University of Pittsburgh, USA and worked at World Bank at Washington, DC. During 1998-1999, she has been a visiting Professor at the Penn State University, USA. She has also been a visiting Scientist at INRIA, France. She has been a principal investigator of many research and consultancy research projects in Neural Networks and Machine learning. She has published a number of research papers in reputed international journals in the area of Neural Networks, Classification, Clustering and Association rule mining. She is also the author of three books. M.P. Gupta is a Professor at the Department of Management Studies, Indian Institute of Technology Delhi (IIT Delhi), India. His research interest lies in the area E-governance that include 15 Doctoral thesis, 11 sponsored projects worth 4$50 million, co-authored book Government Online and about 176 research papers that appeared in National and International Journals/Conference Proceedings. He founded the International Conference on E-governance (ICEG) in 2003. He is the recipient of the prestigious Humanities and Social Sciences (HSS) fellowship of Shastri Indo-Canadian Institute, Calgary (Canada) and was a Visiting Fellow at the University of Manitoba. Professor M.P. Gupta is the corresponding author and can be contacted at: [email protected]

To purchase reprints of this article please e-mail: [email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints