A New Time Series Approach in Churn Prediction with Discriminatory Intervals

Nowadays, the markets have become more competitive. Considering the costs of attracting new customers, organizations want to find churn customers as accurately as possible. In this paper, we propose an approach using time series classification for the churn prediction problem. Most of the previous research uses non-sequential data and doesn’t consider customer behavior in time. Our approach uses a combination of discriminatory intervals and kernel transformers. We use an open-source Retail Transaction dataset. The results show that our algorithm, with sufficient training time, finds the churn customers. We compare our method with other time series classification algorithms; it’s almost 16% higher than the others in the Sensitivity metric. It means that our algorithm finds more customers that truly churn.