A Study Of Mining Passengers’ Mobility Correlation Patterns Based On Big Transit Smart Card Data

As an economical and convenient traffic mode,public transportation has kept attracting a huge amount of passengers to travel by.Meanwhile,ubiquitous social relation bonds trigger passengers to conduct accompanying trips,to finish multiple correlated travel activities,whose aims are commuting,festival rushes,or business cooperation,entertaining,or even pickpocketing crimes,etc.Grasping relation proximity for a variety of passengers in public transit in terms of temporal,spatial,and attributive aspects helps identify multiple mobility correlation patterns in an intelligent way,thus providing guidance to have a visual analytics of their mobility correlation.This study focuses on mining passengers’ mobility correlation patterns in public transportation systems by using big smart card data.We firstly clean the raw data,fill missing key fileds,and construct complete trip chains,and then explore temporal variations in passengers’ activity spaces across a long temporal period.An automatic framework is proposed to detect mobility correlation patterns among abnormal groups in public transportation systems based on passengers’ mobility correlations in time,space,and mobility diversities.Furthermore,a more general model is constructed to examine relation proximity of any pair of passengers by modeling their correlations in profile similarities,interaction activity,and neighborhood-driven interactions.By integrating the aforementioned results,we develop an intuitive,interactive,and efficient visual analytical system to explore mobility patterns of target passengers in public transportation systems based on a “global-to-local” framework.Specially,the thesis falls into the following parts.Firstly,mining temporal variations in activity space based on big transit smart card data.We explore temporal variations in activity spaces for a massive amount of passengers who use public transport based on big transit smart card data generated over a long period of time in Beijing.Passengers are firstly differentiated as having daily routines or not.Their activity spaces are characterized by six features concerning space coverage,distance coverage,and frequency coverage on a per-day transition and with an accumulation of days,respectively.How these features vary over time in these two scenes,as well as how they are related to each other,are statistically analyzed.Results show that the two types of passengers present distint variation patterns regarding days,weeks,and months,which may provide references to characterize abnormal travel behaviors and detect outlier groups in public transportation systems.Secondly,detecting mobility correlation patterns of abnormal groups in public transportation systems based on supervised classifications.Since the current detection of abnormal events in public transportation systems is heavily laborconsuming,we propose a mining approach to automatically detect mobility correlation patterns of abnormal groups.Passengers are firstly characterized by travel patterns regarding temporal,temporal and attributive aspects,and then segmented into several groups based on priori knowledge.Passengers in each group are clustered into multiple subgroups,and then labeled as outliers or not in a hierarchical way.Hence,a sample dataset of outliers is created automatically.The proposed mobility patterns of outliers are then learned based on supervised classification.Meanwhile,the relation network among the outliers is modeled as a graph,whose vertices are denote as discrete outliers,and edge weights are quantified by a combined similarity on mobility pattern,space and time.A molarity-optimum community detection algorithm is adopted to detect outlier groups.Experiments are conducted on pickpocketing gangs reported by SINA microblog data to verify the performance of the proposed method.Thirdly,modeling mobility correlation patterns of passengers in public transportation systems based on a weighted model regarding a joint probability.Apart from the exploration of mobility correlation patterns for a specific group,we further propose an even more general model to explore passengers’ relation proximity.The model measures passengers’ protogenetic relation strength derived from their profile similarities and interaction activities.Meanwhile,it calculates neighborhooddriven relation strength to detect mobility correlation patterns for pairwise passengers in any types.The classification results on a sample dataset containing 4,162 pairs of passengers verify that the model can achieve a higher classification precision by 7.50% compared to several state-of-art approaches.Fourthly,developing a visual analytical system of mining passengers’ mobility correlation patterns based on big transit smart card data.Based on the above research results,we develop an intuitive,interactive and efficient visual analytical system to detect specific targets and their companions in public transportation systems intelligently.Candidates are firstly extracted from fragment travel hints which are automatically integrated by the system.Their mobility correlation patterns are explored in a macro and micro way by employing a “global to local” framework,to help visualize and characterize mobility patterns of potential targets and their companions.Three realworld case studies,as well as performance evaluations given by 30 participants,have demonstrated the effectiveness of the system in detecting,tracking,and characterizing targets.Overall,the research shows how to use big transit smart card with rich practical examples in the transportation field.Findings derived from this study can enhace a deep application of big smart card in transportation security sector.They can provide guidance to satisfy travel demands of passengers featured with multiple types of mobility patterns.Besides,they can provide supports for security departments to detect specific pickpocketing gangs,as well as missing targets in public transportation systems.