Prediction Research Of Online Purchasing Behavior Based On Feature Selection And Model Ensemble

Online shopping has become an indispensable part of people's daily life.In the process of online shopping,customers and businesses do not need to transact face to face,so sellers cannot have a good grasp on the ideas and needs from consumers.But any detail of customers' shopping behavior is recorded in servers,which makes it possible to understand consumers through the analysis of these behavioral data and even to predict their buying behavior.Therefore,this paper proposes to use machine learning,one of the technique of Big Data analysis,to study the purchase pattern behind the data from a large number of consumers' historical online shopping behavior data.When the new shopping behavioral data are input into the model,predicting customer's buying behavior can be realized.After doing a literature review on the influencing factors and prediction of the online buying behavior,we have a deep understanding on the nature of online purchasing behavior and find that the research of the buying behavior based on Big Data is still in its infancy.Therefore,this paper models the online purchase behavior with machine learning algorithms based on the Big Data competition organized by Alibaba.The real shopping behavior data of users from the Alibaba e-commerce platform are used as the research data.Firstly,322 features are constructed from the original data with Sql Server and 10 features which are the most helpful to predict the purchase behavior are selected by using Extra-trees algorithm.Then,Logistic Regression(LR)and Support Vector Machine(SVM),two popular machine learning algorithms,are chose in this paper.And the 10 features are respectively input into the two algorithms to get two prediction models.Finally,the two algorithms are combined with the soft-voting method.Experiments show that the ensemble model is better on prediction than the two single model.This study is driven by data and aims at demonstrating the feasibility of using consumers' historical shopping behavior to predict their future purchase behavior.The forecasting model can be used in the recommendation systems of the shopping websites to realize the complete personalization of users' interfaces,stimulate customers' desire to purchase and improve the conversion rate of-e-commerce platforms.