W Eric Wong Editor Proceedings of the 4th International Conference on Computer Engineering and Networks CENet2014 Proceedings of the 4th International Conference on Computer Engineering and Networks W Eric Wong Editor Proceedings of the 4th International Conference on Computer Engineering and Networks CENet2014 Editor W Eric Wong University of Texas at Dallas Plano, TX, USA ISBN 978-3-319-11103-2 ISBN 978-3-319-11104-9 (eBook) DOI 10.1007/978-3-319-11104-9 Springer Cham Heidelberg New York Dordrecht London Library of Congress Control Number: 2014957309 © Springer International Publishing Switzerland 2015 This work is subject to copyright All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed Exempted from this legal reservation are brief excerpts in connection with reviews or scholarly analysis or material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work Duplication of this publication or parts thereof is permitted only under the provisions of the Copyright Law of the Publisher’s location, in its current version, and permission for use must always be obtained from Springer Permissions for use may be obtained through RightsLink at the Copyright Clearance Center Violations are liable to prosecution under the respective Copyright Law The use of general descriptive names, registered names, trademarks, service marks, etc in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use While the advice and information in this book are believed to be true and accurate at the date of publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made The publisher makes no warranty, express or implied, with respect to the material contained herein Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com) Contents Volume I Part I Algorithm Design A Spatiotemporal Cluster Method for Trajectory Data Yunbo Chen, Hongchu Yu, and Lei Chen Use Case Points Method of Software Size Measurement Based on Fuzzy Inference Yue Xie, Jilian Guo, and Anwei Shen 11 ATPG Algorithm for Crosstalk Delay Faults of High-Speed Interconnection Circuits Yuling Shang and Pei Zhang 19 Application of a Fuzzy-PID Control Method to Synchronized Control of a Multistepping Motor Wenhao Shi, Lu Shi, Fang An, Zhouchang Wu, Zhongxiu Weng, and Lianqing Zhao 27 The Improved Bayesian Algorithm to Spam Filtering Hongling Wang, Gang Zheng, and Yueshun He 37 Evolution of Community Structure in Complex Networks Lei Zhang, Jianyu Li, Shuangwen Chen, and Xin Jin 45 Construction of a Thermal Power Enterprise Environmental Performance Evaluation Model Xiaofei Liao, Huayue Li, Weisha Yan, and Lin Liu 55 A Mobile Localization Algorithm Based on SPSO Algorithm Maoheng Sun and Azhi Tan 65 v vi Contents A Fast and Accurate Algorithm of Subspace Spectrum Peak Search Based on Bisection Method Yu Wang, Hong Jiang, and Donghai Li 73 10 A CRF-Based Method for DDoS Attack Detection Yu Wang, Hong Jiang, Zonghai Liu, and Shiwen Chen 11 Advanced SOM Algorithm Based on Extension Distance and Its Application Haitao Zhang, Binjun Wang, and Guangxuan Chen 89 A Trilateral Centroid Localization and Modification Algorithm for Wireless Sensor Network Yujun Liu and Meng Cai 97 Simulation Study on Trajectory Tracking in Manipulator Based on the Iterative Learning Control Algorithm Yanfen Luo 107 An Improved Gaussian Mixture Model and Its Application Guang Han 115 GPU Acceleration for the Gaussian Elimination in Magnetotelluric Occam Inversion Algorithm Yi Xiao and Yu Liu 123 12 13 14 15 16 17 18 19 20 Artificial Neural Networks in Biomedicine Applications Jiri Krenek, Kamil Kuca, Aneta Bartuskova, Ondrej Krejcar, Petra Maresova, and Vladimir Sobeslav 81 133 SOC Prediction Method of a New Lithium Battery Based on GA-BP Neural Network Kai Guan, Zhiqiang Wei, and Bo Yin 141 Compressed Sensing for Channel State Information (CSI) Feedback in MIMO Broadcast Channels Yuan Liu and Kuixi Chen 155 Implementation and Performance Evaluation of the Fully Enclosed Region Upper Confidence Bound Applied to Trees Algorithm Lin Wu, Ying Li, Chao Deng, Lei Chen, Meiyu Yuan, and Hong Jiang 163 A New Linear Feature Item Weighting Algorithm Shiyuan Tian, Hui Zhao, Guochun Wang, and Kuan Dai 171 Contents 21 22 23 24 25 26 27 28 vii Trust Value of the Role Access Control Model Based on Trust Xiaohui Cheng and Tong Wang 179 Universal Approximation by Generalized Mellin Approximate Identity Neural Networks Saeed Panahian Fard and Zarita Zainuddin 187 Research and Application of Function Optimization Based on Artificial Fish Swarm Algorithm Meiling Shen, Li Li, and Dan Liu 195 Robust Hand Tracker Using Joint Temporal Weighted Histogram Features Zhiqin Zhang, Fei Huang, and Linli Tan 201 Combination of User’s Judging Power and Similarity for Collaborative Recommendation Algorithm Li Zhang, Yuqing Xue, and Shuyan Cao 209 An Improved Naăve Bayes Classier Method in Public Opinion Analysis Yun Lin, Jie Wang, and Rong Zou 219 Overseas Risk Intelligence Monitoring Based on Computer Modeling Peipei Su 227 Quality of Service-Based Particle Swarm Optimization Scheduling in Cloud Computing Shuang Zhao, Xianli Lu, and Xuejun Li 235 Part II Data Processing 29 Improving Database Retrieval Efficiency Shaomin Yue, Wanlong Li, Dong Han, Hui Zhao, and Jinhui Cheng 30 Improving TCP Performance in Satcom Links by Packet-Loss Detection Yuan He, Minli Yao, and Xiong Xiong 31 Key Management Scheme in Cluster for WSNs Xiaoming Liu and Qisheng Zhao 32 An Energy-Saving Method for Erasure-Coded Distributed Storage System Lei Yang and Shi Liu 245 253 263 271 viii 33 Contents LF: A Caching Strategy for Named Data Mobile Ad Hoc Networks Li Zhang, Jiayan Zhao, and Zhenlian Shi 279 34 Topological Characteristics of Class Collaborations Dong Yan and Keyong Wang 35 Cluster Key Scheme Based on Bilinear Pairing for Wireless Sensor Networks Xiaoming Liu and Qisheng Zhao 299 The LDP Protocol Formal Description and Verification Based on CPN Model Rengaowa Sa, Baolier Xilin, Yulan Zhao, and Neimule Menke 305 36 37 38 39 40 41 Self-Adaptive Anomaly Detection Method for Hydropower Unit Vibration Based on Radial Basis Function (RBF) Neural Network Xueli An 323 Design and Implementation of Virtual Experiment System Based on Universal Design Yun Liu, Guoan Zhao, Dayong Gao, and Zengxia Ren 331 Effects of Information Services on Economic Growth in Jilin Province Fang Xia, Bingbing Zhao, and Xiaochun Du 341 Day-Ahead Electricity Demand Forecasting Using a Hybrid Method Zirong Li, Xiaohe Zhang, Yan Li, and Chun Liu 349 Membrane System for Decision-Making Problems Lisha Han, Laisheng Xiang, and Xiyu Liu 43 Operational Model Management C/S System Based on RUP Rui Guo 45 315 A Fast Distribution-Based Clustering Algorithm for Massive Data Xin Xu, Guilin Zhang, and Wei Wu 42 44 291 Decision Analysis Method Based on Improved Bayesian Rough Set and Evidence Theory Under Incomplete Decision System Zhihai Yang, Weihong Yu, Yan Chen, and Taoying Li An Enhanced Entropy-K-Nearest Neighbor Algorithm Based on Attribute Reduction Lingyun Wei, Xiaoli Zhao, and Xiaoguang Zhou 357 365 371 381 Contents 46 47 48 49 50 51 ix Synthetic Safety Analysis: A Systematic Approach in Combination of Fault Tree Analysis and Fuzzy Failure Modes and Effect Analysis Guannan Su, Linpeng Huang, and Xiaoyu Fu 389 Evaluation Model of Internet Service Provider Attraction Based on Gravity Model Lihua Heng, Gang Chen, and Zongmin Wang 399 An Application of Ecological Adaptation Evaluation of Orthoptera in Daqinggou Nature Reserve Using SPSS Chunming Liu, Tao Meng, and Bingzhong Ren 407 Intelligent Diagnostics Applied Technology of Specialized Vehicle Based on Knowledge Reasoning Licai Bi, Yujie Cheng, Dong Hu, and Weimin Lv 415 On the Evaluation of Influence of Golf Websites in China Fangzhi Liu 425 An Ensemble Learning Approach for Improving Drug–Target Interactions Prediction Ru Zhang 433 52 A Complementary Predictor for Collaborative Filtering Min Chen, Wenxin Hu, and Jun Zheng 53 Knowledge Discovery from Knowledge Bases with Higher-Order Logic Guangyuan Li 451 Numerical Analysis on High-Altitude Airdrop Impact Processing of Water Bag Hong Wang, Tao Xu, and Yahong Zhou 459 Heterogeneous Data Sources Synchronization Based on Man-in-the-Middle Attack Yunze Wang and Yinying Li 467 54 55 56 Direct Forecast Method Based on ANN in Network Traffic Prediction Congcong Wang, Gaozu Wang, Xiaoxiao Zhang, and Shuai Zhang Part III 57 443 477 Pattern Recognition Visual Simulation of Three-Point Method Guidance Trajectory for Antitank Missile Mengchun Zhong, Cheng Li, and Hua Li 487 Chapter 147 Modeling for Information Transmission of Consumer Products Quality and Safety Based on the Social Network Yingcheng Xu, Xiaohong Gao, Ming Lei, Huali Cai, and Yong Su Abstract This chapter considers the web information of consumer products quality and safety as research object As to the transmission characteristics of social network, we established an information transmission model without the government’s intervention based on the social network to analyze the relationship of parameters in terms of the information transmission by means of simulation The results show that the proposed model is more effective and feasible Keywords Web information • Consumer products quality safety • Information transmission model • Social network 147.1 Introduction In recent years, incidents in respect of the consumer product’s quality and safety often occur in rapid succession in our country, such as Kumho Tires, Da Vinci furniture, faucet lead beyond the standard, poison uniforms event, and so on; these events directly affect national economy and people’s livelihood They are not only the top priority of consumer attention but also the focus of public opinion and governments With the development of web 2.0, especially the appearance of emerging social media which users are greatly involved in such as WeChat, blogs, wikis, Weibo, BBS, social networks, and content community The modes of information transmission are more and more diversified and complicated Web has gradually become an important platform for the information release of product quality safety, evolution, and monitoring of public opinions Sarafidis carried on further researches on network public transmission in network community and Y Xu • X Gao • H Cai (*) • Y Su Quality Management Branch, China National Institute of Standardization, 100191 Beijing, China e-mail: springblue410@126.com M Lei Quality and Technical Review Center, 710048 Xi’an, China e-mail: kathy-lei@163.com © Springer International Publishing Switzerland 2015 W.E Wong (ed.), Proceedings of the 4th International Conference on Computer Engineering and Networks, DOI 10.1007/978-3-319-11104-9_147 1291 1292 Y Xu et al furthermore proposed scientific warnings and effective intervention strategies [1] Isham et al proposed the correlation functions of degree based on the rumor spreading model and analyzed the influence of network topology on rumor spreading by numerical calculation [2] Becker et al proposed a generic framework which utilized clustering method to identify relevant events of social media [3] Zhang et al discussed the information dissemination model on online social network [4] Xie et al carried out some simulation researches on two competition topics of dissemination based on complex network [5] Wang et al carried out relevant researches on evolution of public opinions of the microblog based on complex network theory [6] Jalili analyzed the formation of social power and public opinions based on complex network as well as carried out simulation experiment in terms of small-world and scale-free network [7] Zhang et al studied the relationship between the tie-strength and information propagation on online social networks [8] Although numerous researches have been carried out, the results are still far from satisfaction Firstly, the depth and the mode of web information dissemination need to be further analyzed; secondly, the information dissemination rule in terms of main channels, structure, path, and evolutionary cycles should be investigated as well On the basis of previous researches, this chapter attempts to build an information transmission model of product quality and safety incidents based on social network with contributions made primarily from the following two aspects, namely, the proposal of a life-cycle model and the media influence factors including the characteristics of product quality and safety information and the establishment of an information transmission model of product quality and safety with the social network 147.2 Information Transmission Mode Based on the Social Network The essence of newly developed social network is online social network service Different from the organization way based on content, online social network consists of users Participating users join in network, release personal information and any content, and connect with anyone related to them As a kind of online service platform, it strives to build and react to social network made up of human relations Online social network is made up of users and connections Imagine there are n members in a social network, of which the ith member is Ai (i ¼ 1, , n), the relation of Ai and Aj is rij, then online social network can be represented as a form of matrix E ¼ (rij)n  n Users often join online social network by means of registration The connections among users can be acquaintance in real world or known based on virtual world According to the difference of website services, connections can be bidirectional or unidirectional Online social network has big advantages in terms of information sharing and seizing social relationship Early online social network can be dated to 147 Modeling for Information Transmission of Consumer Products Quality 1293 the network formatted by e-mail sending-receiving among e-mail users Current online social networks like social network and blog have become the most influenced network media forms r 11 r 21 E¼6 4ÁÁÁ r n1 r 12 r 22 ÁÁÁ ÁÁÁ r n2 ÁÁÁ r 1n r 2n 7 ÁÁÁ5 r nn This section primarily studies information transmission in social network Essentially, human behavior modes can be summarized based on the following two kinds: Push mode (information-diffusing): When node A in the network possesses information and has transmission will, node A will transmit the information to another node B actively Similarly, node A transmits information to a node and node set one by one or at the same time Information transmission acts as radial pattern at this moment Pull mode (information-seeking): When node A in the network possesses information and another node B is interested in the information, node B will emit the curiosity signal when it knows node A possesses the information After that, node A transmits information to node B and the whole information transmission process competes Similarly, this kind of information transmission chain is formed among many nodes Now the network becomes a complex information transmission network 147.3 Information Transmission Model of Consumer Products Quality and Safety In the SIR model, when the susceptible (S) crowd touches infective (I) crowd, they always turn into infective (I) state passively and won’t exit whether they accept to become infective (I) subjectively This is discrepant with the information transmission model based on social network which we are studying; therefore, we propose another model on the basis of SIR model Our study model continuously maintains two characteristics of SIR transmission model Individual transmission process is divided into multiple statuses and infected person is immune ultimately In order to describe the transmission process, we divide the individual statuses into four statuses such as the susceptibly infected status, the non-informed and may be told (S, suspected); the infection status, the informed (I); the involved in transmission (D, disseminating) status, the informed but not transmitted (including stopping transmission status); and R (resistant) status The status of the node in the network continuously changes along with individual decision Status transition of node is shown in Fig 147.1 1294 Y Xu et al =1 Suspected Informed Disseminating Resistant Fig 147.1 Status transition of node We need attention in the model Status D is a child status of status I Actually we can see status I as an intermediate status Upon the node’s arrival at status I, it immediately chooses to be transmitted to status D or not transmitted to status R At the moment of t, the number of nodes in status S is We introduce status I in order to conveniently compute the number of insiders in the model When counting actually, status I, status D, and status R will be computed repeatedly, which needs S(t) + D(t) + R(t) ¼ As shown in Fig 147.1, see from status D at the moment of t The node loses interest in information with probability δ after the transmission’s completion The status turns into the stopping transmission status R When the node is in the susceptible infection status S, D will have two kinds of changes while touching the transmission node One change is interested in information and reading information (inform will), and the node will be transmitted from status S into status I The other change is not interested in information and reading information (no inform will) The node doesn’t change status and maintain at status S When the nodes read information and become informed (status I ), a part of those will produce transmission aspiration after becoming insiders and then will be D The other parts of the nodes that don’t produce transmission will become immune nodes and stop transmission In reality, the number of individual directional contracted people is equal and closed to obey the Poisson distribution; therefore, the network to be discussed is homogenous network Imagine the average degree of network is k, according to status transition of node in Fig 147.1; at the moment of t, we can get the formula as follows: dS=dt ẳ StịDtị > > < dI=dt ẳ StịDtị 147:1ị dD=dt ẳ StịDtị Dtị > > : dR=dt ẳ ịStịDtị ỵ Dtị where I(t) ¼ D(t) + R(t), S(t) + D(t) + R(t) ¼ Parameters in the model are explained as: Relative contract proportion α: The proportion of friends who can accept information appointed by disseminators in all friends The proportion is influenced by the disseminator’s subjective will and network-provided functions In social network and blog, the functions of sharing and forwarding are usually facing all friends with the contract proportion reaching 147 Modeling for Information Transmission of Consumer Products Quality 1295 Inform will η: The probability of the individual who receives disseminator’s information scans the information After receiving the information, the individual will make a rough estimation of information content by scanning title The event attributes title contains (severity of the event, potential impact on users made by event), which will be an important gist for individual estimation By estimating the earnings (inform) and cost (time cost) of scanning the information, the individual will make a decision whether they’ve known each other deeply Therefore, the inform rate will be functioned by both relative contract proportion and inform will Inform rate λ: The proportion of informed friends in all friends when the information disseminators transmit information to friends The proportion is decided by the number of friends selected by such disseminators and the subjective will of scanning the information In this model, the inform rate λ is the product of relative contract proportion α and inform will η; thus λ ¼ α * η In social network and blog, α ¼ 1, so λ ¼ η Transmission will β: The probability of insiders transmitting the scanned web information of consumer quality and safety to friends As an insider, the individual will make a decision whether or not to transmit information to friends The decision is influenced by the user’s character and behavior on the one side On the other side, it is influenced by the earnings (such as satisfaction of the drawing attention) and the time cost acquired by transmitting the information Proportion of losing interest δ: The probability of disseminator losing interest in or forgetting the transmitted information After transmitting information of consumer product’s quality and safety, the individual may believe the information unworthy and thus lose interest in or forget the information Taking α ¼ into differential Eq (147.1), we get Eq (147.2): dS=dt ẳ StịDtị > > < dI=dt ẳ StịDtị dD=dt ẳ StịDtị Dtị > > : dR=dt ẳ ịStịDtị ỵ Dtị 147:2ị where I(t) ¼ D(t) + R(t), S(t) + D(t) + R(t) ¼ 147.4 Experiment Study Simulating the above-described model and imagining there are 100 nodes as SIR model assumption, there are three sickened and infective nodes, 96 non-sickened and infective nodes, and one lifelong immune node at the moment of t ¼ 1; thus S (t) ¼ 0.96, I(t) ¼ 0.04, D(t) ¼ 0.03, and R(t) ¼ 0.01 Setting up parameters η ¼ 0.8, β ¼ 0.5, δ ¼ 0.1, and numerical simulation of 50 steps (in the simulation process compute D(t) and R(t) firstly, and then compute S(t) and I(t)), we get the currency of S(t), I(t), D(t), and R(t) as shown in Fig 147.2: 1296 Y Xu et al Fig 147.2 Changing currency Fig 147.3 I(t) changes with α As can be seen from Fig 147.2, S(t) decreases and R(t) increases when the time t increases as similar as SIR model The newly added node number D(t) will increase and then decrease as the time increases S(t) increases as time increases, but tends to a value less than finally This indicates that all the nodes in the network can’t become insiders even if the whole transmission process finishes, which is different from all the nodes that will be infected in SIR model This is because we add parameter of inform will η Susceptible nodes will choose to be informed or not informed according to their own inform will when contracting information; nevertheless, in the SIR model, they passively become infective On the basis of simple analysis to the model, we know that the inform proportion is the proportion of information-informed persons in all the persons At the moment of t, the inform proportion in the network can be indicated by I(t)/(S(t) + I(t) + R(t)) In the above model, the inform proportion is I(t), which is an increasing process and tends to a specific value when time increases Figure 147.3 shows that the bigger the 147 Modeling for Information Transmission of Consumer Products Quality 1297 Fig 147.4 I0 changes with α α is, the faster the inform proportion will become maximum Figure 147.4 shows the ultimate value of I(t), I ¼ limt!1 I ðtÞ changes as α changes, and the bigger α is, the bigger the ultimate value of I(t) is Conclusion In this study, we propose an information transmission model without the government’s intervention as to the event information characteristics of consumer product’s quality and safety The research on the transmission rule of product’s quality and safety based on social network has provided decision support for the monitoring and management of web information of consumer product’s quality and safety Acknowledgements This research is supported by the National Natural Science Foundation of China (Grant Nos 71301152, 71271013 and 71301011), National Social Science Foundation of China (Grant No 11AZD096), National Key Technology R&D Program of the Ministry of Science and Technology (Grant Nos 2013BAK04B02 and 2013BAK04B04), Quality Inspection Project (Grant No 201410309), and China Postdoctoral Science Foundation (Grants Nos 2013T60091 and 2012M520008) References Sarafidis Y What have you done for me lately-release of information and strategic manipulation of memories Econ J 2007;117(3):307–26 Isham V, Harder S, Nekovee M Stochastic epidemics and rumors on finite random networks Physica A 2010;389:561–76 Becker H, Naaman M, Gravano L Learning similarity metrics for event identification in social media In: Proceedings of the third ACM international conference on Web search and data mining, New York, USA, DBLP, 2010 p 291–300 1298 Y Xu et al Zhang YC, Liu Y, Zhang HF, et al The research of information dissemination model on online social network Acta Phys Sin 2011;60(5):60–6 Xie MS, Jia Z Simulating the spreading of two competing public opinion information on complex network Appl Math 2012;3:1074–8 Wang R, Jin YS, Li F A review of microblogging marketing based on the complex network theory In: 2011 International conference in electrics, communication and automatic control proceedings New York: Springer; 2012 p 1053–60 Jalili M Social power and opinion formation in complex networks Phys A Stat Mech Appl 2013;392(4):959–66 Zhang H, Wang D, Wang L, Bi Z, Chen Y A semantics-based method for clustering of Chinese web search results Enterp Inf Syst 2014;8(1):147–65 Chapter 148 A Multi-classifier-Based Multi-agent Model for Wi-Fi Positioning System Shiping Zhu, Kewen Sun, and Yuanfeng Du Abstract Fingerprint-based Wi-Fi localization systems have become attractive for researchers in indoor location-based services Due to the fluctuant characteristics of received signal strength (RSS) and the lack of the research on environmental factors affecting the signal propagation, the accuracy of the previous systems heavily relies on environmental conditions In this chapter, we propose a novel multi-agent fusion algorithm which combines multiple classifiers Unlike previous multi-classifier combination rule, the proposed approach considers the relativity among classifiers according to co-decision matrix Experimental results show that the multi-classifier approach outperforms single classifier in the test environment with the average accuracy and standard deviations greatly improved in the test environment Keywords Wi-Fi localization • Multi-agent fusion algorithm • Co-decision matrix • Multi-classifier 148.1 Introduction In recent years, the rapid development of mobile smart phone technologies has made indoor location-based service (LBS) more available, such as indoor localization, navigation, and location-based security; nevertheless, owing to the complicacy of the indoor environment, GPS (global positioning system) can’t provide reliable and precise positioning services in indoor environment WLAN (wireless local area networks)-based indoor positioning system has been extensively studied with a lot of solutions proposed in the past two decades [1], and most of the researches have focused on received signal strength indication (RSSI) method Compared with time-of-arrival (TOA) and angle-of-arrival (AOA) [2] algorithms, RSS can be easily received by a Wi-Fi-integrated mobile device S Zhu • K Sun (*) School of Computer and Information Engineering, Hefei University of Technology, 230009 Anhui, China e-mail: kewen.sun@hfut.edu.cn Y Du School of Electronic and Information Engineering, Beihang University, 100000 Beijing, China © Springer International Publishing Switzerland 2015 W.E Wong (ed.), Proceedings of the 4th International Conference on Computer Engineering and Networks, DOI 10.1007/978-3-319-11104-9_148 1299 1300 S Zhu et al According to the published results and literature surveys [3, 4], RSS (fingerprint) methods outperform other techniques in indoor positioning scenarios Fingerprint wireless positioning method [5] is performed on two phases: off-line phase and online phase In off-line phase, RSSs are received at various positions of the target place and stored in a database called radio map In online phase, RSS measured by a WLAN-enabled device is used to estimate a location by means of a variety of techniques, such as k-NN [4], Gaussian distribution [6], Bayesian [7], and PPMCC [8] In this work, we propose a multi-classifier approach for Wi-Fi-based positioning system, and introduce a multi-agent model [9] for multiple classifier fusion In order to demonstrate the effectiveness of multi-classifier, we have evaluated the proposed system in the test environment by adopting three classifiers such as k-NN, Gaussian distribution, and PPMCC There are few studies related to the multi-classifier method for Wi-Fi-based indoor localization One introduced a fuzzy rule-based multi-classification system by using standard methodologies for component classifier generation such as bagging and random subspace along with fuzzy logic to deal with the huge uncertain characteristics of Wi-Fi signal [10]; another combined the Bayesian combination rule and the majority vote for multi-classifier, assuming that the classifiers are independent between each other [11] In our work, we introduce a multi-agent combination approach, which can provide higher positioning accuracy in our experiment; it considers the correlation of each classifier, while the Bayesian rule is not taken into account The remainder of this chapter is organized as follows In Sect 148.2, we introduce a multi-classifier approach for the Wi-Fi-based positioning systems In Sect 148.3, the effectiveness of the proposed system is demonstrated by experiments This chapter summarizes our work and suggests further research in section “Conclusion” 148.2 Proposed Method The algorithm of the proposed system can be illustrated in Fig 148.1 In the off-line phase, the RSSs from all APs (access points) to each RP (reference point) are collected in the test environment as learning data The learning data can be divided into two parts, the classifier training set U1 and the fusion training set U2 The classifier training set U1 is used to construct the fingerprint database Assume that there are m APs, denoted by AP1, AP2, , APm, and n RPs, denoted by RP1, RP2, , RPn Each fingerprint is composed of a pair of data containing MAC address of an AP and its signal strength, and each RP consists of multiple pairs of data, such as {hMAC1, RSS1i, hMAC2, RSS2i, hMAC3, RSS3i, } After collecting the fingerprint, sort the RSSs for each RP and cluster the RPs at the same order of RSSs from detectable APs [12] This cluster not only reduces the computational overhead from Youssef’s research [13], but also reduces the sparsity 148 A Multi-classifier-Based Multi-agent Model for Wi-Fi Positioning System 1301 Fig 148.1 Architecture of proposed indoor positioning system of the confusion matrices, confidence matrices, and co-decision matrices which will be used in the rest of this chapter The fusion training set U2 is used to construct the confusion matrix C In the positioning system, if there are M possible locations, the confusion matrix C contains K M  M matrices C(K ) (k ¼ 1, 2, , K ) and the element of cij(k) represents the number of the samples collected in location i, which is assigned to location j by the classifier k Obviously, the total number of samples collected in location i can be expressed (k) as a row sum ∑ M i ¼ cij , and the total number of samples assigned to location j can (k) be expressed as a column sum ∑ M i ¼ cij Then, the confidence matrix B can be calculated by the confusion matrices C according to Bayes’ theorem, which can be written as below: ðkÞ cij bki ðxÞ ¼ PðEðxÞ ¼ ijek ðxÞ ¼ jk Þ ¼ X M k kị 148:1ị c iẳ1 ijk The element of bki(x) represents the probability that the fingerprint x is assigned to location jk by the classifier k; as to any unknown fingerprint x, the row sum of confidence matrix B(x) is equal to 1, namely, XM iẳ1 bkx xị ẳ k ¼ 1, 2, , K ð148:2Þ The co-decision matrix D shows the decision correlation made by the two 1302 S Zhu et al  à classifiers That is, D ¼ dj1 , j1 , i, k1 , k2 MÂMÂMÂKÂK , where the element of dj1 , j1 , i, k1 , k2 can be defined as below: À Á dj1 , j1 , i, k1 , k2 ¼ P E ¼ iek1 ¼ jk1 , ek2 ẳ jk2 148:3ị The classier k1 assigns ngerprint x to jk1 , the classifier k2 assigns fingerprint x to jk2 , and the fingerprints x belonging to i are the conditions, under which dj1 , j1 , i, k1 , k2 demonstrates the probability A3 dj1 , j1 , i, k1 , k2 ¼ pffiffiffiffiffiffiffiffiffiffi A1 A2 È ẫ A1 ẳ ẩxExị ẳ i, ek1 ẳ jk1 , 8x U ẫ A2 ẳ ẩxExị ẳ i, ek2 ¼ jk2 , 8x U ẫ A3 ẳ xExị ẳ i, ek1 ẳ jk1 , ek2 ¼ jk2 , 8x U ð148:4Þ In (148.4), A1 is the number that the location i is assigned to location jk1 by classifier k1 A2 is the number that the location i is assigned to location jk2 by classifier k2 A3 is the number that the location i is assigned to location jk1 by classifier k1 and also assigned to location jk2 by classifier k2 In online phase, as to the unknown fingerprint x, the final result can be obtained by the following methods: (a) Calculate the confidence matrix B(x) (b) Define the decision matrix Z ¼ [zki]K  M, in which the element of zki is the probability that classifier K assigns fingerprint x to the location i The row sum of Z is equal to 1, and the diagonal elements denote the probability that the classifiers assign fingerprint x to the correct location Initially, assume Z(x) ¼ B(x) (c) The matrix S is theifinal belief value, which can be expressed as below: h X K S ¼ K k¼1 zki , i ¼ 1, 2, , M, and l is the location with maximum 1ÂM belief value Smax (d) If Smax > T, go to (h); otherwise, go to (e) T is a certain threshold (e) Change the decision matrix Z according to the correlation between classifiers, which can be expressed as below: zki ẳ zki ỵ K is K X pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi d j, j1 , i, k, k1 Á zki Á zk1 , i K k ¼1, k 6¼k 1 ð148:5Þ used to adjust the increment within the proper range, does not oscillate, and can converge quickly (f) Normalize each row of matrix Z, still making it to (g) Recalculate Smax and l, and go to (d) (h) Let l be the final location for the multi-classifier 148 A Multi-classifier-Based Multi-agent Model for Wi-Fi Positioning System 148.3 1303 Experimentation 148.3.1 Experimental Setup The experiment is carried out on the 12th floor in the Tian Chuang Technology Building at Zhongguancun in Beijing, where the dimension of the space is 55  10 m Referring to Fig 148.2, there are APs and 25 RPs (19 in the corridor, in the room) The distance between every two neighboring RPs is m In order to collect data, we used the SAMSUNG I9100 mobile phone with Android 2.3.5 platform, and adopted the API provided by the platform For each RP, 120 samples of the fingerprints are collected from each AP 60 samples are used in the training set U1, 30 samples are used in the fusion training set U2 and 30 samples are used in the testing set U3 To analyze the effectiveness of multi-classifier, we use three classifiers, k-NN (with k ¼ 1), Gaussian distribution, and PPMCC to build multi-classifier, and the performance comparison of multi-classifier with these three classifiers, as shown in Table 148.1 When calculating the maximum belief value Smax, set T equal to 0.6 148.3.2 Experimental Results It is clear to observe that the multi-classifier outperforms the conventional single classifier with the positioning accuracy, which is shown in Fig 148.3 The average positioning accuracy of the multi-classifier is 1.55 m, while the average positioning Machine Room Depository Depository Office Meeting Room Office 25 19 18 17 Meeting Room 16 15 Office 14 13 Machine Room 12 11 Machine Room Office 24 10 Office Office 22 23 Office 21 20 R&D Room Fig 148.2 Layout of the floor and positions of the 25 reference points Table 148.1 Average positioning accuracy results for different classifiers Classifiers Average (m) Std dec (m) 67th Percentile 95th Percentile k-NN GAUSS PPMCC Multi-classifier 1.87 1.78 3.16 1.55 3.79 3.84 4.59 3.77 1.5 9 11.24 8.5 1304 S Zhu et al Cumulative Distribution Function Multi-agent KNN PPMCC GAUSS 0.9 0.8 0.7 Probability 0.6 0.5 0.4 0.3 0.2 0.1 0 10 15 20 25 30 35 40 45 Accuracy (m) Fig 148.3 Cumulative distribution function with all positions accuracy values of k-NN, Gaussian distribution, and PPMCC are 1.87, 1.78, and 3.16 m, respectively Table 148.1 illustrates the average positioning accuracy results for different classifiers The standard deviation of the positioning accuracy of multi-classifier is 3.77 m, while the standard deviations of the positioning accuracy for k-NN, Gaussian distribution, and PPMCC are 3.79, 3.84, and 4.59 m, respectively By analyzing the results, we conclude that multi-classifier can improve the accuracy in localization, which can be considered as a promising approach in WiFi-based positioning systems Conclusion In this chapter, the multi-classifier-based multi-agent model is proposed in Wi-Fi-based positioning system The experimental results have proved that the localization performance of the proposed approach outperforms the existing k-NN, Gaussian distribution, and PPMCC in terms of the positioning accuracy and standard deviations In our current work, only three classifiers have been used In further research, we will propose more efficient classifier fusion approach to reduce the complexity overhead by the multiple numbers of classifiers, and more classifiers will be adopted for the multi-classifier fusion in the near future 148 A Multi-classifier-Based Multi-agent Model for Wi-Fi Positioning System 1305 References Deak G, Curran K, Condell J A survey of active and passive indoor localisation systems Comput Commun 2012;35(16):1939–54 Borenovic M, Neskovic A Comparative analysis of RSSI, SNR and noise level parameters applicability for WLAN positioning purposes In: Proceedings of the IEEE EUROCON 2009; 2009 p 1895–900 Farivar R, Wiczer D, Gutierrez A, Campbell RH A statistical study on the impact of wireless signals’ behavior on location estimation accuracy in 802.11 fingerprinting systems Washington, DC: IEEE Computer Society; 2009 p 1, 8, 23–9 Liu H, Darabi H, Banerjee P, Liu J Survey of wireless indoor positioning techniques and systems IEEE Trans Syst Man Cybernet C Appl Rev 2007;37(6):1067–80 Fang S-H, Lin T-N, Lee K-C A novel algorithm for multipath fingerprinting in indoor WLAN environments IEEE Trans Wireless Commun 2008;7(9):3579–88 Youssef MA HORUS: a WLAN-based indoor location determination system College Park: University of Maryland; 2004 Madigan D, Elnahrawy E, Martin R Bayesian indoor positioning systems In: Proceedings of INFOCOM IEEE Computer Society: Washington, DC; 2005 p 1217–27 Tsui AW, Chuang Y-H, Chu H-H Unsupervised learning for solving RSS hardware variance problem in WiFi localization Mobile Netw Appl 2009;14(5):677–91 Kou Z-B, Zhang C-S Multi-agent based classifier combination Chinese J Comput 2003;26 (2):174–9 (in Chinese) 10 Trawinski K, Alonso JM, Herna´ndez N A multiclassifier approach for topology-based WiFi indoor localization Soft Comput 2013;17(10):1817–31 11 Shin J, Jung, SH, Yoon, G, Han, D A multi-classifier approach for WiFi-based positioning system Electrical engineering and applied computing Lecture notes in electrical engineering, LNEE, vol 90; 2011 p 135–47 12 Chen L-H, Wu EH-K, Jin M-H, Chen G-H Homogeneous features utilization to address the device heterogeneity problem in fingerprint localization IEEE Sensors J 2014;14(4):998– 1005 13 Youssef M, Agrawala A, Shankar A WLAN location determination via clustering and probability distributions In: Proceedings of the first IEEE international conference on pervasive computing and communications Washington, DC: IEEE Computer Society; 2003 p 143–50 .. .Proceedings of the 4th International Conference on Computer Engineering and Networks W Eric Wong Editor Proceedings of the 4th International Conference on Computer Engineering and Networks CENet2014. .. 1063716089@qq.com © Springer International Publishing Switzerland 2015 W. E Wong (ed.), Proceedings of the 4th International Conference on Computer Engineering and Networks, DOI 10.1007/978-3-319-11104-9_1... Duplication of this publication or parts thereof is permitted only under the provisions of the Copyright Law of the Publisher’s location, in its current version, and permission for use must always