Research on Customer Churn Prediction Based on PSO-SA Feature Selection Algorithm

The customer churn prediction is a key component in maximizing customer value for businesses, particularly in the highly competitive and customer-centric telecoms industry. The large volume and sparsity of telecoms customer data present significant challenges in churn prediction. To address the issue of exploding data dimensions, the study proposes a feature selection algorithm combining PSO and SA. To evaluate its effectiveness and robustness, the algorithm was trained on multiple predictive models, including Logistic regression, Random Forest, Naive Bayes, and Ensemble models. Comparative evaluation against popular feature selection techniques (e.g., Relief, LVW, LASSO, PSO, SA) demonstrated the superior performance of the proposed algorithm, achieving a highly competitive accuracy of 87%, surpassing other algorithms by 8%. The algorithm proposed in this paper can assist telecoms enterprises in managing customer churn more effectively and enhancing corporate revenue.