Online Learning Customer Churn Prediction Model Based on GA-XGBoost

The online education industry faces a significant challenge with high customer churn rates, making it crucial for companies to effectively predict churn and implement retention strategies. This study develops a customer churn prediction model using an optimized XGBoost algorithm, aiming to identify potential churners in advance and offer actionable solutions to address churn. To optimize the XGBoost model's hyperparameters, five tuning techniques were employed: Random Search, Grid Search., Bayesian Optimization, Gradient Optimization., and Genetic Algorithms. The optimal parameter combination was identified, and the time taken for each tuning process was recorded to assess model efficiency. Experimental results demonstrate that the optimized model performs excellently in F1 score, AUC, and other metrics, significantly enhancing the accuracy of churn predictions. Additionally, the analysis of tuning speeds across different algorithms improves the efficiency of model optimization.