Mobile Phone Changing Prediction based on Large-scale User Behavioral Data

The behavior that people change their cell phones is very important for phone manufactures and mobile operators to make commercial plans and competing strategies to attain maximum benefit. In this paper, we aim to evaluate four existing prediction models in phone changing events, based on user behavior data collected from one telecommunication operator. First we adopt the feature extraction algorithms to distinguish which attributes are closely related with the changing phone label. Secondly undersampling, synthetic minority oversampling technique(SMOTE) and cost-sensitive methods are discussed and implemented. Then four classifiers(i.e. logistic regression, back propagation(BP) neural network, support vector machine(SVM) and random forest(RF)) are compared through extensive experiments and we concluded that BP algorithm in the undersampling scenario can attain better and satisfactory performance.