User Intention Prediction Method Based on Hybrid Feature Selection and Stacking Multi-model Fusion

Abstract

The domestic communication business market tends to be saturated, and the market competition of telecom operators is becoming increasingly fierce. How to mine and predict customers' potential business needs and consumption behavior intentions from users' massive data based on big data is crucial to the fine operation and marketing strategy of telecom operators' existing customers. Due to the complexity of operator network data and the diversity of user behavior, there are many limitations and challenges in the research of user value feature mining, accuracy and generalization of prediction models. In order to solve the above problems, this paper provides a user intention prediction method based on the fusion of mixed feature selection and stacking model. First, based on the hybrid feature selection model of Filter mode and weighted Random Forest, the influencing factors are mined, and the best feature subset is screened; The stacking model fusion framework is proposed, and the FWRF_Stacking hybrid ensemble model based on four classifiers is constructed according to the combination strategy of the model diversity evaluation method and the weighted average method. Finally, it is verified on the real data set of operators. The experimental results show that the prediction model proposed in this paper is superior to other baseline models in multiple performance indicators, and has better effect and applicability for the prediction of telecom customers' business consumption intention.