User Behavior Analysis Based On Vehicle Trajectory Data

With the development of emerging technologies such as the Internet of Things,the mobile Internet,and big data,we are able to access to increasingly information about vehicles driving on the roads.In the field of public transportation,the analysis of trajectory data has achieved good results,but for private cars,the research result is relatively limited.The number of public transportation vehicles only accounts for about 1% of the total number of motor vehicles,which has slight effect on urban traffic planning,mitigation of traffic congestion,traffic guidance,and road condition prediction.Accordingly,it has become an important research issue how to use vehicle driving data to analyze passengers travel behaviors and predict depart time and destinations in the transportation field.Taking Shenzhen as an example,the private cars' GPS data and urban traffic zone data is used to study the user's travel behavior in this paper.The specific research contents and innovation points include:1.Proposing a departure time prediction method based on data mining: First analyze the depart time characteristics and regularity of the user,and then use the K-Means algorithm converts a continuous time variable to discrete time category,starting with Random Forests to predict user next departure time,and uses time-based k-fold cross-validation method and Mutual Information to evaluate the accuracy of model.The joint model based on k-means and Random Forest has an accuracy rate of 75% when the coverage rate is 70%,which is better than 62% of the Naive Bayesian model;2.Buliding a destination prediction model based on Random Forest: This paper applies the Random Forest model to predict the destination,uses time-based k-fold cross-validation method and Mutual Information to evaluate the accuracy of model.When the coverage rate of the model is 70%,the accuracy rate is 75%,which is better than 68% of the Naive Bayesian model.When the coverage rate is 30%,the accuracy rate is 90.48%,which is equal to the 90% accuracy of the Didi prediction system.3.Developing a carpooling system based on the prediction of users' travel behaviors: apply the prediction of users' departure time and destination to the carpool travel system to facilitate users to find carpool vehicles.Easily into practice,this alogrithmn has been applied to BMW's internal carpooling system.Based on the research and analysis method of this paper,the user travel analysis and prediction system can be buit,which is helpful to predict the user's travel characteristics in advance,realizes traffic road condition prediction,traffic guidance,carpool and other applications,to alleviate traffic congestion,further improves road traffic efficiency,and improves the overall operational efficiency and service quality of the transportation system.