Chapter 12
A Multi-Classifier Approach for WiFi-Based Positioning System Jikang Shin, Suk Hoon Jung, Giwan Yoon and Dongsoo Han
Abstract WLAN fingerprint-based positioning systems are a viable solution for estimating the location of mobile stations. Recently, various machine learning techniques have been applied to the WLAN fingerprint-based positioning systems to further enhance their accuracy. Due to the noisy characteristics of RF signals as well as the lack of the study on environmental factors affecting the signal propagation, however, the accuracy of the previously suggested systems seems to have a strong dependence on numerous environmental conditions. In this work, we have developed a multi-classifier for the WLAN fingerprint-based positioning systems employing a combining rule. According to the experiments of the multi-classifier performed in various environments, the combination of the multiple numbers of classifiers could significantly mitigate the environment-dependent characteristics of the classifiers. The performance of the multi-classifier was found to be superior to that of the other single classifiers in all test environments; the average error distances and their standard deviations were much more improved by the multi-classifier in all test environments. J. Shin (&) S. H. Jung Department of Information and Communications Engineering, Korea Advanced Institute of Science and Technology, 373-1 Kusong-Dong, Yuseong-gu, Daejeon, 305-701, Korea e-mail:
[email protected] S. H. Jung e-mail:
[email protected] G. Yoon Department of Electrical Engineering, Korea Advanced Institute of Science and Technology, 373-1 Kusong-Dong, Yuseong-gu, Daejeon, 305-701, Korea e-mail:
[email protected] D. Han Department of Computer Science, Korea Advanced Institute of Science and Technology, 373-1 Kusong-Dong, Yuseong-gu, Daejeon, 305-701, Korea e-mail:
[email protected]
S. I. Ao and L. Gelman (eds.), Electrical Engineering and Applied Computing, Lecture Notes in Electrical Engineering, 90, DOI: 10.1007/978-94-007-1192-1_12, Ó Springer Science+Business Media B.V. 2011
135
136
J. Shin et al.
12.1 Introduction With the explosive proliferation of smart phones, WLAN (Wireless Local Area Network)-based positioning systems have increasingly become a main stream in Location-based Service (LBS) regimes. Compared with other technologies such as GPS [1], RFID [2], GSM [3], Ultrasonic [4], infrared-based systems [5], etc., the WLAN-based positioning systems have some advantages in terms of coverage and costs. Most of the researches on the WLAN-based positioning systems have used the so-called Received Signal Strength Indication (RSSI) from the wireless network access points mainly because the RSSI (or called fingerprint) is relatively easy to obtain using software and also one of the most relevant factors for positioning. Some studies have been reported that consider other factors such as Signal to Noise Ratio (SNR), Angle of Arrival (AOA), and Time of Arrival (TOA) for positioning systems. Milos et al. [6] examined the SNR as an additional input factor and reported that the consideration of both SNR and RSSI could increase the performance of the WLAN-based positioning system. Yamasaki et al. [7] reported that the AOA and TOA are also important factors in positioning. However, the acquisition of the factors including the AOA, TOA, and SNR are not always possible in every wireless network interface cards. Thus, the RSSI appears to have been adopted as a primary factor for the WLAN-based positioning systems. In fact, utilizing the strengths of Radio Frequency (RF) signals for the positioning may not be a simple work. Due to the intrinsic characteristics of the RF signals like multipath fading and interference between signals, the signal strength may severely change depending on the materials used, the positions of doors and windows, the widths of the passages, the numbers of APs deployed, etc. Even if the fundamental parameters are known previously, the derivation of the path loss function of a WLAN signal is extremely complex. In this reason, the WLAN fingerprint-based positioning systems have mostly used to take statistical approaches [6]. The statistical approaches previously suggested have applied various machine learning techniques to derive the positions from the measured fingerprints [2, 8–15]. Those techniques usually are comprised of two phases: off-line and on-line phases. In the off-line phase, fingerprints are captured at various positions of target place and stored in a database called a radio-map. In the on-line phase, the location of a fingerprint is estimated by comparing it with the stored fingerprints in the database. The main problem of the WLAN fingerprint-based positioning systems is that the system performance is too much environment-dependent; in other words, there are not yet any general solutions available for the WLAN fingerprint-based positioning systems. Each system is designed to tackle different environments, and there is no analysis on the relation between the algorithm used and the test
12
A Multi-Classifier Approach for WiFi-Based Positioning System
137
environments. One method may outperform other methods in an environment, but it may show inferior results in other environments. For instance, Youssef et al. [12] suggested a joint-clustering technique and confirmed in their evaluation that their proposed algorithm outperformed RADAR [2]. According to the experiment by Wilson et al. [11], however, the RADAR was found to have a superior performance as compared to the joint-clustering technique. Similarly, this kind of problem was also observed in our experiments. In this paper, we introduce a multi-classifier for the application of the WLAN fingerprint-based positioning systems. We have combined multiple classifiers to become an efficient environment-independent classifier that can realize the more stable and higher estimation accuracy in a variety of the environments. The motivation for using a multiple number of classifiers lies in the fact that the classifier performance is severely environment-dependent; thus, if we can select the most accurate classifier for a given situation, we may be able to achieve even better performance in diverse environments. In this work, a multiple number of classifiers were combined using the Bayesian combination rule [16] and majority vote [17]. To prove the combination effects of the classifiers, we have evaluated the proposed system in three different environments. The evaluation results revealed that the multi-classifier could outperform the single classifiers in terms of the average error distances and their standard deviations. This indicates that the proposed combining method is much more effective in mitigating the environment-sensitive characteristics of the WLANbased positioning systems. The remainder of this paper is organized as follows. The overview on the WLAN fingerprint-based positioning is given in Sect. 12.2. We introduce a multiclassifier for the WLAN fingerprint-based positioning systems in Sect. 12.3. Section 12.4 describes the experiment setup and results. Section 12.5 summarizes this work and suggests the future work.
12.2 Related Work The location estimation using the so-called WLAN fingerprint often refers to the machine learning problem due to the high complexity of the signal propagation estimation. In this reason, various machine learning techniques have been applied. The RADAR system developed by Bahl et al. [2] is considered one of the most representative WLAN fingerprint-based systems. In this system, the authors used the Pentium-based PCs as access points and also the laptop computers as mobile devices. The system uses the nearest neighbor heuristics and triangulation methods to infer a user location. It maintains the radio map which can chart the strength of the signals received from the different access points at some selected locations. Each signal-strength measurement is then compared against the radio map, and
138
J. Shin et al.
then the best matching positions are averaged, enabling the location estimation. Roos et al. [10] proposed the probability-based system which uses the received signal strength samples to create the probability distributions of the signal strength for some known locations. Once an input instance is given, it matches to these probability distributions to find out the location of the mobile device with the highest probability. The histogram method suggested by Castro et al. [18] is another example of the probability-based system. Instead of using Gaussian distribution, it derives the distribution of the signal strength from the learning data. In addition, the adaptive neural networks [13], decision tree [14, 15], and support vector machine [19] are popular on the WLAN-based positioning systems; Kushki et al. [8] suggested the kernelized distance calculation algorithm for the inference of the location of the measured RSSI. Recently, some researchers have focused on compensating the characteristics of the RF signals. Berna et al. [20] suggested the system using the database by considering the unstable factors related to open/close doors and humidity changing environments. They utilized some sensors to capture the current status of the environment. Yin [15] introduced the learning approach based on the temporally updated database in accordance with the current environment situation. Moraes [21] investigated the dynamic RSS mapping architecture. By Wilson Yeung et al. [11], the use of the RSSI was suggested that are transmitted from the mobile devices as an additional input. Thus, there are two types of databases: the RSSI transmitted by APs and the RSSI transmitted by mobile devices. In the on-line phase, the system inferences the multiple results from the databases and makes the final decision using the combining method. Some research efforts [12, 22] have tackled the issue on how to reduce the computational overhead mainly because the client devices are usually small, selfmaintained and stand-alone, having a significant limitation in their power supply. Youssef et al. [12] developed a joint-clustering technique for grouping some locations in order to reduce the computational cost of the system. In this method, a cluster is defined as a set of locations sharing the same set of access points. The location determination process is as follows: for a given RSSI data set, the strongest access points are selectively used to determine one cluster to search the most probable location. Chen et al. [22] suggested the method which selects the most discriminative APs in order to minimize the AP numbers used in the positioning system. This approach selects an appropriate subset of the existing features to the computational complexity problem. Reducing the number of APs is referred to as the dimension reduction in a signal space, which in turn reduces the computational overheads required on the mobile devices. The weak spot of the WLAN fingerprint-based positioning systems is that their performance is severely environment-dependent. One system may outperform the other methods in an environment; it may show an inferior performance in other environments. To solve this problem, we suggest a multi-classifier approach for the application of the WLAN fingerprint-based positioning systems, leading to the more accurate results.
12
A Multi-Classifier Approach for WiFi-Based Positioning System
139
12.3 Proposed Method We utilize the multiple numbers of classifiers using different algorithms to build a possibly environment-independent classifier [23]. The work of combining multiple numbers of classifiers to create a strong classifier has been a well-established research, particularly in the pattern recognition area, the so-called Multiple Classifier System (MCS) [24]. When it comes to the term ‘‘combining’’, it indicates a processing of selecting the most trustable prediction results attained from the classifiers. At least, two reasons may justify the necessity of combining multiple classifiers [25]. First, there are a number of classification algorithms available that were developed from different theories and methodologies for the current pattern recognition applications. For a specific application problem, usually, each one of these classifiers could reach a certain degree of success, but maybe none of them is totally perfect or at least one of them is not so good as expected in practical applications. Second, for a specific recognition problem, there are often many types of features which could be used to represent and recognize some specific patterns. These features are also represented in various diversified forms and it is relatively hard to lump them together for one single classifier to make a decision. As a result, the multiple classifiers are needed to deal with the different features. It also results in a general problem on how to combine those classifiers with different features to yield the improved performance. The location estimation using the WLAN fingerprint often refers to the classification problem because of the noisy characteristics of the RF signals. Many algorithms have been proposed based on the different machine learning techniques, but none of them could achieve the best performance in very diverse environments. At this point, we realized that utilizing the multiple numbers of classifiers could be a promising solution, as a general solution for the WLAN fingerprintbased positioning systems. In this work, we combined the Bayesian combination rule [16] and majority vote [17] for our multi-classifier. The Bayesian combination rule gives weights to the decisions of classifier based on the information in a basis prepared in learning phase. Usually, the basis is given in a form of matrix called a confusion matrix. The confusion matrix is constructed by the cross-validation with learning data in the off-line phase. The majority vote is a simple algorithm, which chooses the one selected by more than a half of the classifiers. Figure 12.1 illustrates the idea of our proposed system. In the off-line phase, the fingerprints are collected over the target environment as learning data. The fingerprint is a collection of the pair-wise data containing the MAC address of an access point and its signal strength. Usually, in one fingerprint, there are multiple tuples of this pair-wise data such as f\ap1 ; bssi1 [; \ap2 ; bssi2 [; \ap3 ; bssi3 [ . . .: g. After attaching the collected location labels to the fingerprints, the database stores the labeled-fingerprint data. After collecting the learning data, each classifier C constructs their own confusion matrix M (Fig. 12.2) using the cross-validation with the learning data. The
140
Fig. 12.1 The overview of multi-classifier
Fig. 12.2 An example of confusion-matrix
J. Shin et al.
12
A Multi-Classifier Approach for WiFi-Based Positioning System
141
confusion matrix would be used as an indicator of its classifier. If there are L possible locations in the positioning system, the M will be a L L matrix in which the entry Mi,j denotes the number of the instances collected in location i, that is assigned as location j by the classifier. From the matrix M, the total number of data collected in location i can be P obtained as a row sum Li¼1 Mi;j , and the total number of data assigned to location P j can be obtained as a column sum Lj¼1 Mi;j When there are K classifiers, there would be K confusion matrices MðkÞ ; 1 k k. In the on-line phase, for the measured Fingerprint x, the positioning results gained by K classifiers are Ck ðxÞ ¼ jk ; 1 k k, and the jk can be any location of the L possible locations. The probability that the decision made by the classifier Ck is correct can be measured as follows: uðjk Þ ¼ Pðx 2 jk jC1 ðxÞ ¼ j1 ; . . .; Ck ðxÞ ¼ jk Þ
ð12:1Þ
Equation 12.1 is called the belief function, and the value of this function is called the belief value. Assuming that all classifiers are independent each other, and applying the Bayes’ theorem to Eq. 12.1, the belief function uðjk Þ can be reformulated as: uðjk Þ ¼
K Y Pðx 2 jk \ Ci ðxÞ ¼ ji Þ i¼1
PðCi ðxÞ ¼ ji Þ
ð12:2Þ
The denominator and numerator in Eq. 12.2 can be calculated using the confusion matrix M. The denominator indicates the probability that the classifier ci will assign the unknown fingerprint x to ji . This can be presented as follows: PL j¼1 Mi;j ð12:3Þ PðCi ðxÞ ¼ ji Þ ¼ P L i;j¼1 Mi;j The numerator in the Eq. 12.2 means the probability that the classifier ci will assign the unknown fingerprint x collected in jk to ji . This term is simply described as below: Mj ;j Pðx 2 jk \ Ci ðxÞ ¼ ji Þ ¼ PL k i i;j¼1 Mi;j
ð12:4Þ
After applying Eqs. 12.3 and 12.4 to Eq. 12.2, Eq. 12.2 can be reformulated as: uðjk Þ ¼
K Y
M P L jk ;ji j¼1 Mi;j i¼1
If more than a half of estimation of the classifiers pointed a specific location, the location would be selected as the final result. Otherwise, the belief value of each prediction is calculated, and the location with the highest belief value would be the final result. In case there are many locations with the same highest belief value, the
142
J. Shin et al.
multi-classifier system determines the middle point of those locations as the final result. For example, assume that there are three classifiers, a, b, and c, and there are three possible locations, location 1, location 2 and location 3. After the off-line phase, the confusion matrices will be as follows: 0 1 18 4 7 B C MðaÞ ¼ @ 2 12 3 A 0 4 10 0 1 12 6 6 B C MðbÞ ¼ @ 3 9 3 A 2 5 11 0 1 14 2 2 B C MðcÞ ¼ @ 4 11 5 A 2 7 13 If the classifiers a, b, and c assigned the unknown instance x to location 1, location 2, and location 3, respectively, the belief values of the predictions can be calculated as follows: 18 3 2 108 ¼ 29 15 22 9570 4 9 7 252 ¼ uðjb Þ ¼ 29 15 22 9570 7 3 13 273 ¼ uðjc Þ ¼ 29 15 22 9570 uðja Þ ¼
The multi-classifier assigns the location 3 to the unknown instance x, because the jc , the prediction of the classifier c, has the highest belief value.
12.4 Evaluation 12.4.1 Experimental Setup The performance of WLAN-based positioning systems depends on each environment itself where the evaluation is performed. In this reason, we evaluated the proposed multi-classifier in three different environments; Table 12.1 briefly illustrates the test environments. The testbed 1 implies an office environment; the dimension of the corridor in the office is 3 9 60 m. The office is on the third floor of the faculty building at the KAIST-ICC in Daejeon, South Korea. In the corridor, we have collected 100 samples of Fingerprints from 60 different locations. Each location is 1 m away from each other. The testbed 2 indicates another office
12
A Multi-Classifier Approach for WiFi-Based Positioning System
Table 12.1 Summary of testbeds
Type Dimension (m) Number of AP Distance between RP (m) Number of APs deployed Avg. number of APs in one sample Std.Dev of number of AP in sample
143
Testbed 1
Testbed 2
Testbed 3
Corridor 3 9 60 60 1 48 16.6
Corridor 4 9 45 45 1 69 16.8
Hall 15 9 15 25 3 36 13.9
1.89
4.24
3.48
environment where the dimension of the corridor is 4 9 45 m. The office is located on the second floor of the Truth building at the KAIST-ICC. We have collected 100 samples of the Fingerprint from 45 different locations. Each location is 1 m away from each other. The testbed 3 implies a large and empty space inside the building located at the first floor of the Lecture building at the KAIST-ICC. The dimension of the space is 15 9 15 m. In the testbed 3, we have collected 100 samples of the Fingerprints from 25 different locations. Each location is 3 m away from each other. Comparing the testbed 3 case with testbed 1 and 2 cases, there is no attenuation factors that may disturb any signal propagation. To collect the data, we have adopted the HTC-G1 mobile phone with Android 1.6 platform, and used the API provided by the platform. We have also used the half (50%) of the collected data as the learning data and the rest of data were used as the test data. To prove the better performance of the multi-classifier, we created the multi-classifier with three classifiers, k-NN (with k ¼ 3) [2], Bayesian [9], and Histogram classifiers [10]; the performance of the multi-classifier was compared with these three classifiers, as shown in Table 12.2.
12.4.2 Results We can observe from the results that none of the single classifier outperformed others in all three test environments. These results indicate that the performance of the WLAN fingerprint-based positioning systems is sensitively related to the environments and the multi-classifier is turned out to be much more effective in mitigating such characteristics of the WLAN signals. Figure 12.3 reports the average error distance with respect to the different numbers of APs. From the Fig. 12.3a and b, the performances of the classifiers are quite different according to the test environments. Although the testbed 1 and testbed 2 look similar each other in indoor environments, the performances in testbed 1 are better than those in testbed 2. Especially, the average error distance of k-NN classifier in testbed 1 was 1.21 m when 15 APs were used for positioning,
144
J. Shin et al.
Table 12.2 Summary of testbeds (meter) Avg Testbed 1
Testbed 2
Testbed 3
k-NN Histogram Bayesian Multi k-NN Histogram Bayesian Multi k-NN Histogram Bayesian Multi
4.0 2.6 2.8 2.4 1.3 2.0 1.3 1.1 4.8 5.8 5.6 4.5
Std.Dev
Max
Min
90th Percentile
5.3 3.8 3.9 3.6 3.0 2.5 1.8 1.6 4.5 4.6 5.1 4.5
43 29 25 25 44 26 17 13 22.5 22.5 22.5 22.5
0 0 0 0 0 0 0 0 0 0 0 0
12.0 7.0 7.0 7.0 3.0 5.0 3.0 3.0 18.03 20.62 21.21 18.03
Fig. 12.3 Average error distance versus number of AP used for positioning in a Testbed 1, b Testbed 2, and c Testbed 3 respectively
while it was 4.6 m in testbed 2. In case of the histogram classifier, the average error distances were 1.9 and 2.7 m with 15 APs in testbed 1 and testbed 2, respectively. With the same condition, the Naïve Bayesian classifier’s average error distances in the testbeds 1 and 2 were 1.25 and 2.47 m, respectively. Compared with other classifiers, the multi-classifier showed the more improved results. In the testbeds 1 and 2, the average error distances of the multi-classifier
12
A Multi-Classifier Approach for WiFi-Based Positioning System
145
with 15 APs were 1.1 and 2.3 m, respectively. In the testbed 3, the accuracies of all classifiers are extremely poorer than the results in other testbeds. Based on the findings, it is believed that the WLAN fingerprint-based positioning systems can show better performance in the office environments as compared to the hall environments involving a few attenuation factors. As shown in the Fig. 12.3, the multi-classifier may clearly mitigate the environment-dependent characteristics of the single classifier. From the results shown in Fig. 12.3, we can conclude that the multi-classifier is effective for reducing error distance in localization. Table 12.2 illustrates the performance summary of the classifiers. The standard deviation of the errors of the multi-classifier in the testbed 1 was 3.6 m, while the k-NN, Histogram, and Bayesian respectively showed 5.8, 3.8, and 3.9 m in their standard deviations. In the testbed 2, the standard deviations of the error of all classifiers were lower than the values in the testbed 1. The standard deviation of kNN, Histogram and Bayesian were 3.0, 2.5, and 1.8 m, respectively. The standard deviation of the error of the k-NN, histogram, and Bayesian classifier in testbed 3 were 4.5, 4.6, and 5.1 m, respectively. These results confirm that the standard deviation of the errors of WLAN fingerprint-based positioning systems is also dependent on the environments. The proposed multi-classifier outperformed others in all testbeds in terms of the standard deviations of the error. In testbed 1, 2, and 3, the standard deviations of the errors of the multi-classifiers were 3.6, 1.6, and 4.5, respectively, which are higher or equivalent performance compared with others. From the results, we confirmed that multi-classifier could mitigate the environment-dependent characteristics of the single classifier, and the performance of the multi-classifier was better than the others in all environments. Even if the improvement of performance was not remarkable, the results indicate that combining a number of classifiers is one of the promising approaches in constructing reliable and accurate WLAN fingerprint-based positioning systems.
12.5 Summary and Future Work In this paper, we have presented an environment-independent multi-classifier for the WLAN fingerprint-based positioning systems in an effort to mitigate the undesirable environmental effects and factors. We have developed a combining method of the multiple numbers of classifiers for the purpose of the error-correction. For example, if a single classifier predicted wrong, the other classifiers correct it. In other words, the classifiers in the multi-classifier can complement each other. We have evaluated the multi-classifier in three different environments with various environmental factors: the numbers of APs, the widths of corridor, the materials used, etc. The multi-classifier was constructed with three different classifiers; k-NN (with k ¼ 3), Bayesian, and Histogram classifiers. As a result, the multi-classifier showed a consistent performance in the diverse test environments while other classifiers showed an inconsistent performance. The performance of
146
J. Shin et al.
the multi-classifier tends to follow that of the single classifier showing the best performance. This means that the classifiers in the multi-classifier complement each other, and thus the errors are more effectively corrected. For the next step, we are going to investigate a more efficient combining rule. In this work, we have mixed the Bayesian combining rule and majority vote; however, the performance enhancement was too marginal. Considering the complexity overhead of using the multiple numbers of classifiers, the multi-classifier may not be a cost-effective approach. Finding the best combination of the classifiers will be another direction of our future work. We have tested only three classifiers, and two of them have taken similar approaches; the fingerprint is the only feature for positioning. There are numbers of systems considering various aspects of WLAN signals that use additional features. In the near future, we are going to implement and evaluate the multi-classifier with various types of classifiers. Acknowledgments This research was supported by the MKE(The Ministry of Knowledge Economy), Korea, under the ITRC(Information Technology Research Center) support program supervised by the NIPA(National IT Industry Promotion Agency) (NIPA-2010-(C1090-10110013)), and by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MEST) (No. 2008-0061123).
References 1. Enge P, Misra P (1999) Special issue on global positioning system. Proc IEEE 87(1):3–15 2. Bahl P, Padmanabhan V (2000) RADAR: an in-building RF-based user location and tracking system. Proc IEEE Infocom 2:775–784 3. Drane C, Macnaughtan M, Scott C (1998) Positioning GSM telephones. IEEE Commun Mag 36(4):46–54 4. Priyantha N, Chakraborty A, Balakrishnan H (2000) The cricket location-support system. In: Proceedings of the 6th Annual International Conference on Mobile Computing and Networking, pp 32–43 5. Want R, Hopper A, Falcão V, Gibbons J (1992) The active badge location system. ACM Trans Inf Syst (TOIS) 10(1):102 6. Borenovic M, Neskovic A (2009) Comparative analysis of RSSI, SNR and noise level parameters applicability for WLAN positioning purposes. In: Proceedings of the IEEEEUROCON, pp 1895–1900 7. Yamasaki R, Ogino A, Tamaki T, Uta T, Matsuzawa N, Kato T (2005) TDOA location system for IEEE 802.11 b WLAN. In: Proceedings of IEEE. WCNC’05, pp 2338–2343 8. Kushki A, Plataniotis K, Venetsanopoulos A (2007) Kernel-based positioning in wireless local area networks. IEEE Trans Mobile Comput 6(6):689–705 9. Madigan D, Elnahrawy E, Martin R (2005) Bayesian indoor positioning systems. In: Proceedings of INFOCOM, pp 1217–1227 10. Roos T, Myllymaki P, Tirri H, Misikangas P, Sievanen J (2002) A probabilistic approach to WLAN user location estimation. Int J Wirel Inf Netw 9(3):155–164 11. Yeung W, Ng J (2007) Wireless LAN positioning based on received signal strength from mobile device and access points. In: IEEE International Conference on Embedded and RealTime Computing Systems and Applications, pp 131–137
12
A Multi-Classifier Approach for WiFi-Based Positioning System
147
12. Youssef M, Agrawala A, Shankar A (2003) WLAN location determination via clustering and probability distributions. In: Proceedings of the First IEEE International Conference on Pervasive Computing and Communications, p 143 13. Borenovi M, Nekovic A, Budimir D (2009) Cascade-connected ANN structures for indoor WLAN positioning. Intell Data Eng Autom Learning-IDEAL 392–399 14. Chen Y, Yang Q, Yin J, Chai X (2006) Power-efficient access-point selection for indoor location estimation. IEEE Trans Knowl Data Eng 18(7):877–888 15. Yin J, Yang Q, Ni L (2008) Learning adaptive temporal radio maps for signal-strength-based location estimation. IEEE Trans Mobile Comput 7(7):869–883 16. Xu L, Krzyzak A, Suen C (1992) Methods of combining multiple classifiers and their application to hand writing recognition. IEEE Trans Syst Man Cybern 22:418–435 17. Kuncheva L (2001) Decision templates for multiple classifier fusion: an experimental comparison. Pattern Recogn 34(2):299–314 18. Castro P, Chiu P, Kremenek T, Muntz R (2001) A probabilistic room location service for wireless networked environments. In: Proceeding of the 3rd International Conference on Ubiquitous Computing, pp 18–34 19. Brunato M, Battiti R (2005) Statistical learning theory for location fingerprinting in wireless LANs. Comput Netw 47(6):825–845 20. Berna M, Lisien B, Sellner B, Gordon G, Pfenning F, Thrun S (2003) A learning algorithm for localizing people based on wireless signal strength that uses labeled and unlabeled data. In: Proceedings of IJCAI, pp 1427–1428 21. Moraes L, Nunes B (2006) Calibration-free WLAN location system based on dynamic mapping of signal strength. In: Proceedings of the 4th ACM International Workshop on Mobility Management and Wireless Access, pp 92–99 22. Chen Y, Yin J, Chai X, Yang Q (2006) Power efficient access-point selection for indoor location estimation. IEEE Trans Knowl Data Eng 1(18):878–888 23. Shin J, Han D (2010) Multi-classifier for WLAN fingerprint-based positioning system. Lecture notes in engineering and computer science: Proceedings of the World Congress on Engineering, WCE 2010, 30 June–2 July, London, UK, pp 768–773 24. Kittler J (1998) Combining classifiers: a theoretical framework. Pattern Anal Appl 1(1):18–27 25. Chen K, Wang L, Chi H (1997) Method of combining multiple classifiers with different features and their applications to text-independent speaker identification. Int J Pattern Recognit Artif Intell 11(3):417–445
