Document Type : Original Article
Authors
1
Department of Computer Science, Faculty of Computer and Information Science, Ain Shams University, and (National Bank of Egypt), Cairo, Egypt
2
Faculty of Computer and Information sciences, Ain Shams University
3
Department Computer Science, Faculty of Computer and Information Sciences,Ain Shams University, Cairo, Egypt.
Abstract
Customer retention is a top priority for organizations due to its significant impact on corporate profitability. There is a lot of competition between banks to acquire and retain customers. The youth customer segment is the future of digital banks, and hence, this study was conducted to forecast the youth segment loyalty. This will help banks identify the degree of customer loyalty and the factors that affect their satisfaction. Customer churn may lead to a financial loss of revenue and market share. Therefore, forecasting customer loyalty has become essential to maintaining profitability and the customer base. Using Fintech (financial technology) and digital transformation techniques in digital banking works on enhancing the youth customers experience and increasing their lifetime value using machine learning techniques. This research presents a new model of stacking ensemble learning, which combines optimized base learner algorithms after applying hyperparameter tuning and the voting model to the stacking meta-learner algorithm. The research compares various base machine learning models, such as KNN (K-Nearest Neighbors), LR (Logistic Regression), RF (Random Forest), Adaboost, and GB (Gradient Boosting), for customer loyalty prediction. The experiment was generated using 10,000 banking customers, which contains 6,420 youth customers. The model assessment proved that using base learners combined with a voting mechanism as an input to stacking modeling received an accuracy of 88.9%. This research discusses challenges related to existing classification models, including mitigating biases and errors, preventing overfitting, addressing imbalanced data, enhancing model stability, improving interpretability, and automating model selection by using hybrid models for tuning.
Keywords
)
View on SCiNiTO