Mobile User Trajectory And Profile Mining

Recent development in smart devices like Smart phones, Ipad, IPod, PDA’s and portable wireless sensors network devices made the mobility data a valuable research field nowadays, where the mobility data can be used for the development of location based services (LBS) like Early warning system, Route prediction, City wide sensing, Mobile marketing, Social networking thorough mobility profile building against user’s mobility history. This mobility profile represents user’s behavioral patterns, spatiotemporal trends; stay points visit history, habits and context, where stay point information is the basic entity to understand user’s trends. Because of its low cost coverage and easy availability GSM Cell Global Identity (CGI) is most appropriate method to be used for user mobility profiling.In this research work we focused mainly on the precise mobility profile building thorough trajectory and behavioral pattern mining using the GSM CGI Cell-ID, where all the concerned issues like precise spatial extraction, stay points detection and mobility profiling are addressed properly thorough the proposed framework. The proposed framework utilized both spatiotemporal and semantic information for mobility profile building which is not addressed in any of previous related work, so makes it suitable for any LBS due to of its novel and generic nature. Followings are the main original contributions of our work:1. Spatial clustering based data pre-processing approach:Data pre-processing done in this work includes extraction of spatial coordinate, removal of spatial outliers and retrieval of missing values, where spatial coordinate extraction is done using Google APIs with partial GSM CGI header information available in used MIT Reality Mining dataset thorough reverse engineering. Spatial outliers detection and retrieval of missing values from the extracted mobility data is done using the proposed Location Area Code (LAC) based spatial clustering and Semantic clustering techniques where data is aged out due to of change in GSM network architecture and rapid shift towards3G network. The experiments on real world dataset show that our proposed clustering techniques based on basic GSM network architecture are effective and efficient to address the related issues.2. Stay points extraction based on user movement correlation:Extraction of stay point is carried out in two steps i.e. cell oscillation resolution and extraction of actual stay points. In first step cell oscillation resolution is done using the basic GSM network features and semantic information provided by user. And in second step user stay points extraction as per spatiotemporal information is done thorough implementation of proposed Geo-grid clustering approach based on overlapping area prediction and dwell time during user movement. To evaluate our proposed methodology we carried out the experiments on real world user mobility data i.e. MIT Reality Mining dataset and successfully extracted the stay points where precision value is81.70%after cell oscillation resolution.3. Semi-supervised frequent pattern mining based user mobility profiling approach:User trend analysis and profile building is done thorough frequent pattern mining approach based on prefix scan algorithm and spatiotemporal information binding, which is not only valuable in user profile building thorough user mobility pattern extraction but also equally valuable for user context prediction. User similarity mining is done by the proposed methodology based on Longest Common Sub Sequence and Co-location properties mined in similarity matrix. To determine the authenticity of our proposed methodology we carried out the experiments on the real world dataset i.e. MIT Reality Mining dataset. Further we compared our hybrid approach that uses both spatiotemporal and semantic information with two other techniques i.e. Spatial Cosine Similarity (SCS) and Extra-role Co-location rate (ERCR) defined over two metrics i.e. Mean Average Precision (MAP) and Normalized Discounted Cumulative Gain (nDCG), the experimental results shows that our proposed methodology outperformed.