User Network Purchase Behavior Prediction Based On Selective Ensemble

The development of Internet technology is getting faster and faster,and as a result,more and more users are shopping online,and the massive amount of commodity information and user information has also caused the problem of information overload.For users,there are too many types of goods and it is difficult to choose.For the platform,how to analyze the user’s psychology from the massive data and realize the prediction of the user’s online purchase behavior has also become a difficult problem.Based on this background,this thesis conducts research on the prediction of users’ online purchase behavior.This thesis studies the prediction method of users’ online purchase behavior based on machine learning,and conducts research on data filling,feature engineering,prediction model establishment and optimization of prediction model based on ensemble pruning.Firstly,aiming at the serious problem of missing key data in the process of data preprocessing,a method of filling online shopping data of users based on random forest method is studied.The random forest method is used to construct multiple decision trees to complete the filling,avoiding the loss of effective information.Then,for the problem that the feature dimension of the dataset is too small,the XGBoost method is used to calculate the feature importance score after the feature group is constructed,and the feature screening is completed,and a large number of effective features are obtained to improve the data dimension,so as to facilitate the mining of the potential features of the data.Secondly,in order to realize the prediction of users’ online shopping behavior,a prediction model is built based on massive online shopping behavior data by using different machine learning methods.At the same time,in order to obtain a model with more stable and accurate prediction performance,a prediction model of users’ online purchase behavior is built based on the method of Stacking ensemble learning,and the effectiveness of the method is verified by experiments.In addition,in order to optimize the prediction model of users’ online purchase behavior,improve the prediction accuracy of the model and reduce the ensemble scale,a selective ensemble method based on probability multi-dimension is studied.On the traditional stacking ensemble method,the ensemble pruning is carried out based on the probabilistic multi-dimensional method,and a prediction model of user online purchase behavior with smaller ensemble scale and higher prediction accuracy is obtained.