User Churn Model in E-Commerce Retail

Abstract

In e-commerce retail, maintaining a healthy customer base through retention management is necessary. Churn prediction efforts support the goal of retention and rely upon dependent and independent characteristics. Unfortunately, there does not appear to be a consensus regarding a user churn model. Thus, our goal is to propose a model based on a traditional and new set of attributes and explore its properties using auxiliary evaluation. Individual variable importance is assessed using the best performing modeling pipelines and a permutation procedure. In addition, we estimate the effects on the performance and quality of a feature set using an original technique based on importance ranking and information retrieval. The performance benchmark reveals satisfying pipelines utilizing LR, SVM-RBF, and GBM learners. The solutions rely profoundly on traditional recency and frequency aspects of user behavior. Interestingly, SVM-RBF and GBM exploit the potential of more subtle elements describing user preferences or date-time behavioural patterns. The collected evidence may also aid business decision-making associated with churn prediction efforts, e.g., retention campaign design.

In e-commerce retail, maintaining a healthy customer base through retention management is necessary. Churn prediction efforts support the goal of retention and rely upon dependent and independent characteristics. Unfortunately, there does not appear to be a consensus regarding a user churn model. Thus, our goal is to propose a model based on a traditional and new set of attributes and explore its properties using auxiliary evaluation. Individual variable importance is assessed using the best performing modeling pipelines and a permutation procedure. In addition, we estimate the effects on the performance and quality of a feature set using an original technique based on importance ranking and information retrieval. The performance benchmark reveals satisfying pipelines utilizing LR, SVM-RBF, and GBM learners. The solutions rely profoundly on traditional recency and frequency aspects of user behavior. Interestingly, SVM-RBF and GBM exploit the potential of more subtle elements describing user preferences or date-time behavioural patterns. The collected evidence may also aid business decision-making associated with churn prediction efforts, e.g., retention campaign design.