Reinforcement Learning in Customer Churn Prediction for Insurance datasets

Predicting customer attrition, a critical challenge for businesses today, presents formidable obstacles. The departure of customers directly impacts an organization's reputation, financial stability, and future growth prospects. Customer behavior can change for various reasons, including uncontrollable factors or unexpected events, leading to shifts in data patterns. This can hamper the predictive accuracy of classifiers developed using traditional supervised learning techniques, especially when they rely on passive learning methods. In response to these challenges, this paper proposes the application of reinforcement learning in the context of customer churn prediction. Specifically, it utilizes the Deep Q Network (DQN) model, tailored for learning on a dedicated customer churn dataset for classification tasks. To adapt to shifting patterns, a distinct dataset is generated using shuffle sampling based on distribution. The evaluation of classifier performance is based on four key metrics: Precision, Recall, F1 score, and Accuracy. In comparison to traditional classifiers like XGBoost, Random Forest, and K-Nearest Neighbors (KNN), the findings underscore DQN's superiority as an active learner. DQN demonstrates enhanced predictive capabilities, especially as the dataset size grows and evolves with the emergence of new patterns.