Dynamic customer behavior prediction in subscription services using causal reinforcement learning

This study addresses a critical gap in predictive analytics for subscription-based services by integrating causal inference with a Modified Markov Decision Process model, enhanced by reinforcement learning, to predict customer behavior dynamically. While traditional approaches in big data analytics focus on customer segmentation and churn probability, they often fail to make adaptable, forward-looking predictions. By identifying causal variables and defining customer states, this research advances segmentation accuracy and captures the evolving value of a customer over time. The proposed methodology, which computes transition probability matrices and initial state probabilities, demonstrates superior predictive accuracy, achieving mean absolute error of 4.27 and root mean squared error of 4.24 on the wikipedia web traffic dataset, and mean absolute error of 10.89 and root mean squared error of 4.25 on a multinational organization dataset. These results signify the efficacy of the proposed method in outperforming existing models, offering valuable insights into the dynamic nature of customer behavior in subscription services.