The 11th Asia Pacific Industrial Engineering and Management Systems Conference The 14th Asia Pacific Regional Meeting of International Foundation for Production Research Melaka, 7 – 10 December 2010
Design of the Optimum Layout in Supermarkets with Application of the Multi-Agent Simulator Masumi Hanada†, Hidetaka Izumo Graduate School of Natural Science, Kanazawa University Kanazawa, Japan
[email protected]† Jun Nakajima FSAS NETWORK SOLUTIONS INC. 6-81 Ogami, Naka-ku, Yokohama, 231-0015, Japan Koji Abe School of Science and Engineering, Kinki University 3-4-1 Kowakae, Higashi-osaka, 577-8502, Japan Takuya Tajima Fukuoka Institute of Technology 3-30-1 Wajiro-higashi, Higashi-ku, Fukuoka, 811-0295, Japan Takehiko Abe Kanazawa Institute of Technology 7-1 Ohgigaoka, Nonoichi, Ishikawa, 921-8501, Japan Hidetaka Nambo and Haruhiko Kimura Graduate School of Natural Science, Kanazawa University Kakuma, Kanazawa, Japan
Abstract - This paper presents a method for designing the optimum layout in supermarkets. From the idea that increase of sales per head brings increase of gross sales in a supermarket and sales per head depends on the length of traffic line, some methods for extending traffic line were reported. Because it is commonly believed that long traffic line increases sales per head, then gross sales will increase in the same way. However, when traffic line is long, sometimes there is possibility the number of visits to the supermarket is decreased due to fatigue of walking. For the reason, this paper proposes a method for designing the optimum layout for increasing sales per head from correlation between articles for sale. The proposed method does not depend on the length in designing the layout. In the experiment for the proposed method, association rules were extracted from point of sale system (POS) data of a supermarket by association analysis and the proposed layout was designed according to the association rules which we obtained in this study. Efficiency of the proposed method has been shown by experimental results that sales per head was increased in a virtual supermarket on the multi-agent simulator. Keywords: association rules, POS, supermarket, flow planning, LTV
1. INTRODUCTION There is a commonly-held view that traffic line extending is required to increase sales value within the retail industry. Expanded traffic line has long been
________________________________________ † : Corresponding Author
recognized as one of the key conditions to improve sales value. For this reason, a variety of extending traffic line techniques has been applied in many stores. However, when traffic line is long, sometimes there is possibility the number of visits to the supermarket is decreased due to
The 11th Asia Pacific Industrial Engineering and Management Systems Conference The 14th Asia Pacific Regional Meeting of International Foundation for Production Research Melaka, 7 – 10 December 2010 fatigue of walking. The aim of this research is proposing a method for designing the optimum layout for increasing sales value per head from correlation between articles for sale. The proposed method does not depend on the length in designing the layout. In the experiment for the proposed method, association rules were extracted from point of sale system (POS) data of a supermarket by association analysis and the proposed layout was designed according to the association rules which we obtained in this study. Efficiency of the proposed method has been shown by experimental results that sales per head was increased in a virtual supermarket on the multi-agent simulator.
Ck+1=Apriori-Gen(Fk); for all transactions t∈D do begin C’t=subset(C`k+1, t); for all candidate c∈C’t do c.count++; end Fk+1={c∈c k+1|c.count≧minimum support} end Answer=Uk Fk;
2. SUPERMARKETS AND DATA MINING
Association rule mining is one of Data mining techniques. A consultant discovers that disposable diapers and beer are frequently bought together. This is the most famous example of association rule mining. association rule mining is useful to discover regularities between items. For example, {bread} => {milk} means that if a shopper buys bread, the shopper have a tendency to buy milk. Thus, we employed association rule mining to analyze POS data. Table 1 shows purchase ratios for the articles obtained from receipt data for 1 year in a supermarket. From Table 1, we could find out that most of the customers purchased 1~10 articles.
2.1 Present Situation of Japanese Supermarkets In 2008, Lehman Brothers Holdings Inc filed for Chapter 11 bankruptcy-court protection because of the subprime mortgage crisis. The bankruptcy caused global financial crisis and recession, and Japan was no exception. It is called “Lehman Shock”. According to the latest Bank of Japan confidence survey, the Japanese manufacturing industry grew for the first time in 2 years since Lehman Shock. On the other hand, non-manufacturing industry is still marking time. Supermarkets belong to the category of non-manufacturing industry. This adverse business conditions hit supermarkets hard. A new study released by Ministry of Economy, Trade and Industry of Japan (METI) indicates that supermarket sales value have now fallen for 11 consecutive months in Japan. In addition to this, we must not ignore the effect of convenience stores (CVS) power on supermarkets. The proportion of sales value of convenience stores in retail industry increased from 5.5% in 2007 to 5.9% in 2009. Traditionally, a big population of shoppers stock up on food and other everyday goods in a supermarket at periodic intervals by car. Due to CVS was higher-priced store in the first place. Originally, CVS supplemented supermarkets with convenience. Today, however, CVS reduced the price differential between supermarkets by private brand (PB). As a result, supermarkets declined in importance of retail industry. We may, therefore, reasonably conclude that a big population of shoppers attached importance to convenience of store in shopping. Still, supermarkets have the advantage of large selection.
2.2 Based Algorithm Algorithm Apriori F1 = (Freequent itemsets of cardinality=1); for (k=1; Fk ≠φ;k++);
This is adapted from [3].
2.3 Association rule mining
Table 1: Purchase ratios for the articles
3. SIMULATOR DEVELOPMENT It is not realistic way to find out an appropriate layout changing layout frequently since the frequent changes need an enormous amount of money and cause an enormous loss. Besides, they just give customers confusion. Hence, we constructed a virtual supermarket using a multi-agent simulator development environment. The name of the development environment is "artisoc". The virtual supermarket was modeled on a supermarket in Kanazawa
The 11th Asia Pacific Industrial Engineering and Management Systems Conference The 14th Asia Pacific Regional Meeting of International Foundation for Production Research Melaka, 7 – 10 December 2010 city, Japan. The virtual supermarket is composed of the store's map, item-agents, and customer-agents. Customeragents represent customers in the supermarket. Item-agents represent articles in the supermarket. Customer-agents enter from the gates on the map into the supermarket. The number of item-agents (i.e., articles) to purchase depends on the customer-agents. As the initial value of the number, a random number between 1 and 10 is given to the customer-agents. After a customer-agent has entered the supermarket, the agent moves grids on the map one by one toward to the location where the item-agent to purchase is put. The location is determined by the ratios shown in Table 1. After the customer-agent reached the location, the agent adds the item-agent into its basket. Next, the customer-agent purchases item-agents which gives rise to its purchase with association rules obtained from Table 2. The customer-agent repeats this behavior until the behavior times become the given initial number.
support ratio are listed in a table (Table 3). Here, association rules that the number of transactions is less than 3000 or lift value is less than 1 are omitted. By displaying the couple of class A and class B which has high support ratio in the neighborhood, the opportunities articles in class A are purchased with articles in class B could be increased. Table 3: Association rules which have a high confidence ratio between the primary classes
Table 2: Association rules which have a high confidence ratio between the articles
4.2 Association rules between the articles
4. ANALYSIS We reached characteristics presented by association rules on purchases in a supermarket from receipts. The supermarket is located in Kanazawa city (a country city in Japan) and the supermarket’s company expands chain stores in the city.
4.1 Association rules between primary classes of articles in supermarket Association rule: regarding A→B, support ratio, confidence ratio, lift value, and transactions which include both of A and B are output in descending order of support ratio. And, association rules whose support ratio is high and the
Association rule: regarding X → Y, support ratio, confidence ratio, lift value, and transactions which include both of X and Y are output in descending order of support ratio. And, association rules whose support ratio is high and the support ratio are listed in a table (Table 2). Here, association rules that the number of transactions is less than 30 or lift value is less than 1 are omitted. By displaying the couple of article X and article Y which has high support ratio in the neighborhood, the opportunities both of X and Y are purchased together could be increased.
5. CONSIDERATIONS We find out other characteristics on the supermarket from the receipts used in 4. And, we consider beneficial information for supermarkets extracted by the characteristics.
The 11th Asia Pacific Industrial Engineering and Management Systems Conference The 14th Asia Pacific Regional Meeting of International Foundation for Production Research Melaka, 7 – 10 December 2010
5.1 Examination items as the characteristics (1) The number of sales for each article: Listing articles in descending order of sales number per a month, make a graph (x-axis: articles, y-axis: the number of sales). And then, list articles whose sales number is large and articles whose sales number is small in a table. (2) Articles whose sales volume is large: Calculate sales volume P(Xi) for article Xi. And, listing the articles in descending order of the volume, make a graph (x-axis: articles, y-axis: sales volume). And then, list articles whose sales volume is large in a table. P(Xi) = price number where price is the unit price for Xi, and number is the sales number for Xi per a month. (3) Transition of sales for the articles whose sales volume is large: Examine the transition of sales per a month for the articles whose sales volume is large.
volume is large between the chain stores, there is possibility of increasing the number and volume for the articles in chain stores where the number and volume are small. Hence, it is necessary to find up the differences between chain stores. (5) Control articles in stock according to the transition of sales number for the articles: By comparing results of (3) in 5.1 between chain stores and finding tendencies of sales, the stores can estimate the number of sales and control articles in stock. (6) Estimate the total sales volume: The supermarket can estimate the total sales volume from results of (4) in 5.1 for each chain store. (7) Estimate the number of cashiers: The supermarket can estimate the number of essential cashiers at each time zone on each day of the week from results of (5) and (6) in 5.1.
REFERENCES (4) Transition of total sales volumes: Examine the transition of total sales volumes per a month. (5) Transition of the transactions in days of the week: (6) Transition of the transactions in unit time:
5.2 Beneficial information for supermarkets (1) Think articles whose sales number is large are important: Generally, regarding articles whose sales number is large, since the possibility customers purchase is high, we can guess it could be easier to sell the articles. Hence, emphasize the articles putting at a place where it is easy for customers to get them, advertising the articles with announcements, and decorating the place.
P. Underhill: Why We Buy: The Science Of Shopping, Simon & Schuster (2004). T. Tajima, T. Abe, H. Kimura: A Simulator for Customers’ Buying Behavior Adopting Correlation of Merchandise, Journal of the Japan Society for Production Management, Vol.16, No.2, pp.241-248 (2010). H. Motoda, S. Tsumoto, T. Yamaguchi and M. Onuma,: Learning Data Mining, Ohmsha, pp.43-46 (2006). Ministry of Economy, Trade and Industry of Japan (2010) Commercial Statistics for May. http://www.meti.go.jp/statistics/tyo/syoudou/result/kakuho_ 1.html. Access date: 7 July 2010 at 15.00.
(2) Neglect articles whose sales number is small: Regarding articles whose sales number is small, since the possibility customers purchase is low, decrease the number of the articles in stock and put at a minor place.
ACKNOWLEGMENT
(3) Think articles whose sales volume is large are important: The reason is the same as (1).
AUTHOR BIOGRAPHIES
(4) Check difference of articles whose sales number and volume are large between chain stores Conducting (1) and (2) in 5.1 for each chain store. Regarding articles that difference of sales number and
This research was supported by KOZO KEIKAKU EN GINEERING Inc (artisoc academic lend-lease contract).
Masumi Hanada received M.E. from Shinshu University. Currently, he is in doctor course at Kanazawa University. His interests include knowledge discovery in databases (KDD) and collective intelligence (CI). He is a member of the IEICE of Japan, and DBSJ (The Database Society of Japan).
The 11th Asia Pacific Industrial Engineering and Management Systems Conference The 14th Asia Pacific Regional Meeting of International Foundation for Production Research Hidetaka Izumo is a student of the Kanazawa University graduate school of natural science and technology in Japan. He graduated from Kanazawa Institute of Technology in Japan in 2010. His research interest is management information. Jun Nakajima received the Bachelor of Engineering from Kanazawa University in Japan in 2008. He is working in FSAS Network Solutions Inc. Koji Abe received his B.S. and M.S. degrees from Kogakuin University, Japan, in 1996 and 1998, respectively. And, he received his Ph.D. degree in Engineering from Kanazawa University, Japan, in 2001. After that, he affiliated in the Institute for Image Data Research, University of Northumbria at Newcastle, UK, as an honorary research fellow in 2002. He was an assistant professor in the Dept. of Info. and Comp. Sci., Kanazawa Institute of Technology, Japan, in 2003. Currently, he is an Associate Professor in the Dept. of Informatics, School of Science and Engineering, Kinki University, Japan. His research interests include pattern recognition, medical image processing, CBIR, multimedia database, and artificial intelligence. He is a member of IEEE, IEICE (Japan), IEEJ (Japan), and IPSJ (Japan). Takuya Tajima received a master's degree from Graduate School of Managerial Engineering, Kanazawa Institute of Technology in 2003. Also in 2003, he joined FSAS Network Solutions Inc. In 2008, he completed his doctoral course in the Graduate School of Natural Science and Technology, Kanazawa University, Japan. He holds a PhD in Engineering. He was a research associate in Department of Electronics and Information Engineering, Ishikawa National College of Technology, Japan in 2007. Currently, he is an assistant professor in Fukuoka Institute of Technology. He is a member of the Japanese Society for Artificial Intelligence (JSAI), Japan Industrial Management Association (JIMA) and JSPM. Takehiko Abe received his BA and PhD degrees from Kanazawa University, Japan in 1988 and 1997 respectively. He joined Daiwa Institute of Research Ltd. in 1988. He is currently a professor at Kanazawa Institute of Technology, Japan. His research interests include data mining and management information. He is a member of the Institute of Electronics, Information and Communication Engineers (IEICE), JSAI, JSPM and the Japan Society for Management Information (JASMIN). Hidetaka Nambo received his B.E. degree in 1994 from Dept. Electrical Inf., Kanazawa Univ., and his M.E and Ph.D. degrees in 1996 and 1999 from Kanazawa Univ. He was a Research Associate in 1999 in the Dept. Electrical
Melaka, 7 – 10 December 2010 Inf., Fac. Eng., Kanazawa Univ. Since 2001, he is a lecturer of Graduate School of Natural Science and Technology, Kanazawa University. He is engaged in research on the user authentication algorithm of pet robots and data mining. He is a member of the IEE of Japan, the IEICE of Japan, the IPSJ of Japan, and the IEEE. Haruhiko Kimura graduated in 1974 with a specialty in Applied Physics and Chemistry from the Engineering Department at Tokyo Denki University. In 1979, he completed his doctoral course in Information Engineering in the Graduate School of Engineering at Tohoku University. He holds a PhD in Engineering. Also in 1979, he joined Fujitsu Corporation. In 1980, he became a Lecturer in Kanazawa Women's Junior College. In 1984, he became an Assistant Professor in the Economics Department of Kanazawa University. Currently, He is Professor in the Information Systems Engineering Department on the Engineering Faculty of Kanazawa University. At this time, he is engaged in research on optimal code conversion and the acceleration of production systems. He is a member of the Instituted of Electronics, Information and Communication Engineers (IEICE), the Japanese Society for Artificial Intelligence and the Information Processing Society of Japan.