Customer Churn Prediction in Banking Sector Using Machine Learning

Abstract

Bank time deposits provide consistent returns on investment. However, there are difficulties in locating and luring new clients. By using a combination of XGBoost, ADSYN, and Random Search optimization strategies to handle data imbalance, this study improves the predictive power of deposit categorization models. The study makes use of a Bank Marketing dataset from the UCI Repository that is openly accessible. comprising 45,211 items with a notable class disparity (88.3% of "no" responses and 11.7% of "yes" responses). While Random Search effectively optimizes model parameters, ADASYN integration enhances minority class representation. Our suggested hybrid model outperforms conventional methods by achieving an accuracy of 94.93%, precision of 94.93%, recall of 94.95%, and ROC-AUC score of 0.9919. These results demonstrate the efficacy of our approach in comparison to baseline models. This hybrid model accomplishes our research goals and improves customer data analysis. We go over the difficulties of integration, such as the need for computation and the choice of methods. The findings highlight the significance of assessing statistical significance in model enhancements and mitigating noise caused by synthetic samples. The study highlights how machine learning can be used to solve problems in the financial sector, with a focus on how feature engineering and data pretreatment affect performance. In order to improve model scalability and further minimize complexity, future research may investigate AutoML, which could lead to more creative consumer data analysis