Online Commodity Purchase Prediction Model Based On Voting Strategy

Online consumption is no longer a rare phenomenon,the entire shopping era has transitioned to the smart era.In offline shopping,consumers and merchants conduct face-to-face transactions,and merchants can directly observe consumers'shopping needs and shopping behaviors.In order to predict the future purchase behavior of users,e-commerce platforms use powerful servers to record and manage online shopping behaviors,and then dig out relevant information from these massive user behavior data to understand online consumers'shopping tendencies.Therefore,how to use data mining algorithms and machine learning algorithms to dig out online users'shopping tendencies from massive user shopping behavior data has become a research hotspot today.First of all,this thesis introduces the research background,significance and current research status at home and abroad,and gives the research framework and method ideas of this thesis.Secondly,the basic theory of the data mining algorithm in the recall phase is introduced,and part of the recalled user-product list is displayed.Third,this article selects real user shopping behavior data(user information data,product information data,and user behavior information data)provided in the Alibaba Cloud competition,and based on the exploratory analysis of the data and the actual business background,to recall the users-items list is the basis for feature extraction,89 data features are constructed,and then the top 30 data features are selected based on the SVM-RFE feature selection algorithm.After the above analysis,the logistic regression model and the random forest algorithm model were selected to construct the user purchase prediction model in this paper.Soft Voting algorithm fusion strategy was used to carry out the algorithm fusion of the above two optimal models to build a mixed model for empirical research.In this process,the main empirical studies include:firstly,an exploratory analysis of users'historical shopping behavior data,such as analyzing which shopping behaviors will most affect future user purchasing behaviors,and analyzing whether the timing of the shopping behaviors affects the future shopping behavior,etc.,and correlate the original three data.Second,in the data exploration and analysis,it is found that the time when the user's shopping behavior occurred in the past has a significant impact on the user's purchase behavior in the future,so the user's interest in product is constructed after considering the time factorR_u,i,and added to the data.Third,the F1 value is used as the evaluation index of the user purchase prediction model,and the hyperparameters of the prediction model are tuned,and the optimal user purchase prediction model is selected through5-fold cross-validation.The empirical results show that the F1 value of the hybrid model constructed based on the Soft Voting algorithm fusion strategy in the training set and the test set is higher than the single machine learning algorithm model,and the prediction model is purchased for the user.Finally,summarize the research process of this thesis,and look forward to the future work.