Research On The Prediction Of Repeated Purchase Behavior Based On Machine Learning Algorithm

Repeated purchase behavior is a research hotspot in the field of marketing.With the rapid development of e-commerce,more and more users participate in online shopping.How to predict the repeated purchase behavior of users based on big data has become a very concerned problem for e-commerce platform.Repeated Purchase Behavior Prediction technology can be applied to the e-commerce platform recommendation system to help merchants identify users with repeated purchase intention,so as to achieve the accurate delivery of marketing information.The key to accurately predict repeated purchase behavior is to excavate the user behavior law implied in the data by meachine learning algorithm.However,because of the large scale of the online shopping user group and the different law of purchase behavior for various user groups,this makes the data mining work become extremely difficult.The traditional machine learning algorithm ignores the difference of the law for user purchase behavior when predicting,and it is difficult to obtain good prediction effect.Therefore,this paper focuses on how to improve the generalization performance of machine learning model in the problem of repeated purchase behavior prediction,so that it can overcome the influence of difference user behavior patterns on the predictive performance.Based on the research of the existing machine learning algorithm,this paper puts forward a fine differentiation ensemble learning method.Based on the research of the existing machine learning algorithm,this paper puts forward a fine differentiation integrated learning method.This method can learn a variety of user behavior laws from the data set,and improve the predictive performance of the model.The main research work of this thesis is as follows.1.Research on the influencing factors of repeated purchase behavior.Through the analysis of the purchase behavior data of the e-commerce platform users,this paper excavates the important factors that affect the repetitive purchase behavior,and constructs 52 kinds of characteristics related to the repetitive purchase behavior.2.A comparative study of classical machine learning models.Several classical machine learning methods are studied in this paper.The experimental results show that the prediction accuracy of the current machine learning model is generally not high.A single model,such as Logistic regression,neural network and decision tree,has limitations in solving the problem of sample category imbalance.Although the ensemble learning method can solve the category imbalance problem by the way of non-sampling,it is not possible to learn the different user behavior law effectively,and the prediction effect is also relatively low.3.Research on fine differentiation ensemble learning.Considering the limitation of the current machine learning method in predicting the repeated purchase behavior,this paper improves the ensemble learning method of Bagging,and puts forward the fine differentiation ensemble learning.Fine differentiation ensemble learning enables the subdivision of user purchase behavior data by setting a strong rule binding strategy in Bagging so that it can filter the sample data layers by layer.This new integrated learning method can learn a variety of purchasing behavior rules based on the subdivided dataset.Experimental results show that the ensemble learning of fine differentiation has better predictive effect than the current machine learning methods.4.Research on generalization performance of predictive models.The generalization performance of the fine differentiation ensemble learning model is studied from the point of view of generalization error.By decomposing the generalization error of the model,it is found that the sample filtering mechanism of fine differentiation ensemble learning can reduce the variance,so it has better predictive performance.