Research on User Portrait Based on Xgboost and Logistic Ensemble Learning Methods

Abstract

In the era of big data, accurately analyzing users' purchasing behavior and specifying strategies for corresponding enterprises can not only reduce enterprise costs but also improve users' consumption habits, which is of great significance to the development of enterprises. In view of this, this paper analyzes the data of user behavior. Firstly, the original data are preprocessed, and then feature engineering is carried out based on correlation. The important features were selected as reference and 15 features with practical significance were extracted. Considering the imbalanced data, this paper adopts resampling to balance the data. After continuous experiments, a combined model based on three models is proposed: Based on the parallel combination of XGBoost and RandomForestClassifier, serial modeling is carried out with LogisticRegression, and the final model is established to predict user consumption behavior. It can be observed from the experimental results that the combinatorial model not only has the advantages of high classification accuracy in tree model, but also has the advantages of strong interpretation and high stability in logistic regression. Due to the great difference between the logic regression model and the tree model, the correlation of the results is low. Therefore, this method is feasible and scientific.