Optimizing Machine Learning Model Selection for Customer Churn Prediction Using Multi-Criteria Decision-Making Methods

In today's market competition scene, it's crucial for marketers to predict customer churn to improve customer loyalty and boost profits. This research delves into selecting machine learning models for forecasting customer churn using multi-criteria decision-making techniques. Using credit card customer churn as an example, we illustrate how various machine learning algorithms can be assessed and prioritized effectively using performance measurements that match marketing goals. We implemented four widely used models-CART, CART with cross-validation, Random Forest, and Logistic Regression-on a real-world dataset, applying data balancing techniques such as over-sampling and under-sampling to address class imbalance. To optimize model selection, we utilized the Analytic Hierarchy Process (AHP) to weigh performance criteria based on expert judgments and the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) to rank the models accordingly. Our findings reveal that the Random Forest model trained on over-sampled data outperforms others, achieving high accuracy, precision, and recall, making it a robust tool for marketers to identify at-risk customers. The proposed approach not only assists in selecting the most suitable predictive model but is also adaptable to various industries beyond credit card services, where customer churn prediction is crucial. By optimizing model selection through MCDM methods, marketers can better tailor their strategies to proactively engage with customers likely to churn, thereby enhancing customer loyalty and increasing revenue.

کلیدواژه ها:

customer churn ، customer retention ، model selection ، MCDM ، machine learning