Enhancing Customer Personality Prediction using Advanced Machine Learning Techniques and Data Balancing Strategies: A Comprehensive Approach to Addressing Imbalanced Datasets in Marketing Analytics

This study tackles imbalanced customer response prediction by integrating advanced resampling methods with diverse machine learning and ensemble techniques. Novel resampling approaches (SMOTE, Borderline-SMOTE, ADASYN, and SMOTEENN) were systematically applied alongside Logistic Regression, Random Forest, Gradient Boosting, LightGBM, SVM, and Neural Networks. Stacking and voting ensembles further consolidated individual model strengths. Results show that coupling resampling with ensembles significantly boosts minority class metrics. For example, ADASYN combined with LightGBM achieved a 95% accuracy and substantially improved minority class recall and precision, while SMOTEENN with a voting ensemble reached a 99% recall, albeit with a slight precision trade-off. These findings underscore that employing sophisticated data rebalancing techniques in conjunction with hybrid ensembles can surpass traditional models, offering a balanced improvement in metrics such as accuracy, precision, recall, and F1 score. The practical significance lies in enabling more reliable predictions in highly imbalanced scenarios, ultimately enhancing decision-making processes.