A data mining-based model for predicting the repurchase behavior of fitness service users

Abstract

This study aimed to develop a predictive model for fitness service reuse using data mining techniques. Four modeling approaches were evaluated: neural networks, decision trees, association rules, and logistic regression. Among them, the neural network model demonstrated the highest predictive performance (Accuracy = 89.7%, Precision = 86.3%, Recall = 88.0%), making it suitable for integration with automated CRM systems and personalized marketing efforts. The decision tree model, while slightly less accurate (Accuracy = 81.2%), offered high interpretability, facilitating strategic decision-making and effective communication with management stakeholders. The association rule model, employing the Apriori algorithm, successfully uncovered behavioral patterns related to weight training and class participation, which can inform the development of targeted promotions and customized service offerings. Logistic regression, known for its simplicity and transparency, was found to be less capable of modeling complex, non-linear relationships within the dataset. The study concludes that a hybrid approach combining neural networks and association rule mining offers complementary advantages high predictive accuracy and actionable insights into user behavior. This integrated method supports the delivery of personalized services, reduces member attrition, and enhances customer retention. Overall, the findings offer practical value for strategic planning in the fitness industry by enabling data-driven decisions to optimize customer engagement and long-term loyalty.