Behavior Analysis And Prediction Of Car-sharing User

The generation and development of the Car-sharing mode is a great achievement of the sharing economy reflected in the way of transportation,which can improve the utilization rate of vehicles,relieve the tension of public resources,reduce the travel cost of users,and is conducive to energy conservation,emission reduction and environmental governance.However,due to the late start of the Car-sharing model in China,it's operators are not accurate enough to understand the demands of users,resulting in inadequate publicity and coupon distribution.In this context,the key to the long-term development of the Car-sharing industry is to predict the user categories and clarify the user behavior in advance,so as to stimulate the user consumption.The sustainability of the Car-sharing user's use will directly affect the scale of user groups and the company's revenue,which is the focus of operators' attention.To clarify the user categories and their use behaviors in advance will help operators to provide accurate and high-quality services for users and carry out efficient customer management.Therefore,this paper constructs a dynamic model to predict the category of user loyalty,and on the basis of discussing the characteristics of user time dimension,constructs a user travel time prediction model,which provides a reference for the scientific management of Car-sharing operators.Based on the actual operation data of Gansu Yixiangxing New Energy Development Co.,Ltd,this paper makes an in-depth study on the behavior characteristics of Car-sharing users.This paper defines two attributes to measure the time when users enter the Car-sharing market and the loyalty of users,the latent ratio and the persistence ratio,which are used as classified input variables to build K-means user clustering model.Combined with the evaluation indicator DBI,users are divided into four categories: Lost users,early loyal users,late loyal users and motivated users.At the same time,this paper creatively constructs a dynamic user category prediction model.By sliding the time nodes of observation period and judgment period,it compares and analyzes the prediction accuracy under different time intervals,so as to achieve the goal of obtaining a small amount of historical data in a short observation period and achieving long-term prediction in a long judgment period.The results show that the six-month observation period can effectively learn user behavior characteristics,and the prediction accuracy of user categories can reach 85%.Based on the results of user segmentation,this paper carries out a comparative prediction research on the travel time of different categories of users.First of all,Fisher ordered clustering method is used to divide a day's time axis into dawn rising period,daytime flat peak period,afternoon peak period,night flat peak period,late night falling period and early morning low period,so as to convert user travel time points into category variables and lay a data foundation for the construction of user travel time point model.Secondly,user key attributes are extracted as input variables by calculating feature importance,and then the travel time prediction model of users is constructed.Finally,the accuracy of the travel time model of all users is 96.59%.Compared with the prediction accuracy,it is found that the prediction effect is better for each type of users to build models separately.The user category prediction model and user travel time prediction model established in this paper have the characteristics of high accuracy and strong practicability,which can provide effective decision support for the dynamic resource management of Car-sharing operators.