RFID Path Data Clustering Analysis And Frequent Pattern Mining

With RFID technologies and applications developing, the data generated by RFID applications is growing rapidly. How to extract useful information and knowledge from large amounts of data has became an important research topic. RFID path data refer to the path data generated by RFID-tagged objects in their movement process. In this thesis, we analyzed the characteristics of RFID data path based on real applications, proposed some clustering analysis approaches and frequent path mining approaches, which suitable for RFID path data. In applications, these approaches can help business decision-making, optimize or improve the business arrangements, support the planning and so on.In cluster analysis, the similarity measurement is the basis of clustering algorithms. We refered to some related alignment techniques in Bioinformatics, discussed two path similarity calculation methods– global and local similarity based, which can reveal the true similarity of paths. The traditional clustering algorithms can not process RFID path data. This thesis presented a density based clustering algorithm DBPC for RFID path data clustering. Based on the characteristics of path data, DBPC proposed a new way to build clusters. We also proposed a hierarchical clustering algorithm PHC for path data, which use a weight scheme to calculate cluster similarity. Finally, we discuss outlier detection methods and complex path data clustering methods.In frequent path mining, we modified the closed frequent sequence mining algorithm CloSpan to make it suitable for path data. Based on CloSpan, we proposed an algorithm CFPM for frequent path mining, which well correspond with characteristics of the path data, use a node counting scheme for tree cutting, and has better performance. Finally we disscuss frequent pattern mining for complex path data.We developed an experiment system for RFID path data mining, which has path visualization functions, can intuitively reveal the distribution of path data and mining results. Experiments show that the mining approaches proposed by this thesis can be well applied to RFID path data. DBPC and PHC can build high quality clusters and have good efficiency. CFPM improves the mining efficiency compare to tranditional CloSpan.