Machine Learning-Based Merchant Churn Prediction in Banking

In markets with service models such as contractual banking, and merchants, new customer acquisition is much more expensive than customer retention. Banks should focus on various marketing activities by detecting the potential of losing customers by using data mining and machine learning methods. In this study, in a sample of a private bank in Turkey, the probability of leaving the next month is calculated by the customer loss analysis called the Churn Analysis of the member businesses determined by the brand in the retail sector and similar parameters. In this context, analysis was made with 128.721 samples and various performance metrics (Accuracy, Precision, Sensitivity, Area Under the Line, and Gini) in the data set created to calculate the separation probabilities. By using 75 different parameters such as transaction volume, number of transactions, number of products in the bank, number of complaints, size of work in the bank, current pricing, and dues payment order of the workplaces in the last 1 year, the leaving score and wallet share decrease scores were calculated for the workplaces. Decision Tree, Random Forest, Cat Boost, XGBoost and LightGBM machine learning algorithms were used for this research. Due to its performance, the LightGBM algorithm was chosen.