Lads: Deep Survival Analysis for Churn Prediction Analysis in the Contract User Domain

Abstract

Survival analysis methods are currently used in the fields of medicine, economics, biology and engineering, and focus on the relationship between covariates and the timing of events. Survival analysis makes such use of survival time that it is better suited to the user domain than other machine learning models, In order to mine information about the user's time-series characteristics, improve the ability of neural networks to process information and the efficiency of network training with large data sets. Based on deep survival analysis, the Lads (LSTM Attention Deep Survival) model was designed to combine Long Short-Term Memory networks and Attention mechanisms to predict the occurrence of events of interest in the contract user domain. The LSTM acts as a feature extractor and performs pre-processing of time-series characteristics information, while the Attention mechanism mainly enhances the interpretability of the model. The final experimental results show that the Lads survival analysis model is a better predictor in the contract user domain than survival analysis methods such as CPH (Cox Proportional Hazard Model) and DeepSurv.