Machine Learning Approaches for Customer Churn Prediction in the Aquaculture Technology Sector

This study investigates the application of advanced machine learning techniques for customer churn prediction in the rapidly evolving aquaculture technology sector. We employ and compare three distinct models—Logistic Regression, Random Forest, and XGBoost—to analyze a synthesized dataset representative of the industry. The research encompasses comprehensive data preprocessing, feature engineering, and model evaluation using standard performance metrics. Our findings demonstrate the superior performance of XGBoost, achieving 88% accuracy in predicting customer churn. Through feature importance analysis, we identify key churn predictors, with the difference between a customer’s last order amount and their mean order amount emerging as the most significant factor. Additionally, we utilize SHAP (SHapley Additive exPlanations) analysis to interpret model outcomes, revealing nuanced relationships between features and churn probability. The study highlights the critical role of consistent engagement, proactive customer support, and personalized retention strategies in reducing churn. Our research contributes to the growing body of knowledge on churn prediction in specialized technology sectors and provides actionable insights for improving customer retention strategies in the aquaculture industry. The paper concludes with recommendations for future research, including the integration of external data sources and exploration of deep learning approaches for temporal dependency analysis in customer behaviour.