Accurately forecasting potential customers, reducing resource waste, and improving efficiency may all be necessary for telemarketing to be productive. To find clients, we use machine learning models such as Random Forest, XGBoost, AdaBoost, and Gradient Boosting. F1 score, recall, accuracy, and precision are used to assess these algorithms. To address class imbalance in telemarketing datasets, oversampling techniques like SMOTE, ADASYN, and Borderline SMOTE generate synthetic data for underrepresented classes. Model performance is improved by MinMax scaling data normalization and Mutual Information feature selection. Training and testing are conducted using telemarketing data from the Portuguese Bank Marketing Data Set. Modifying the XGBoost hyperparameters raises the model's accuracy to more than 98% for both the top 10 features and the complete set. To promote user involvement, the system has a real-time prediction interface built on Flask. Ensemble-based online machine learning, a successful telemarketing optimization technique, enables models to adjust to shifting customer preferences.
Keywords : Ensemble Learning; Online Machine Learning; Telemarketing; Class Imbalance; Synthetic Oversampling; SMOTE; ADASYN; Borderline SMOTE; Mutual Information; Feature Selection; Data Normalization; XGBoost; Random Forest; Gradient Boosting; Real-Time Prediction; Flask Interface; Bank Marketing Data
Author : SK. Himambasha1 , P. Sai Ram2
Title : Improved Ensemble-Based Online Machine Learning Approach for Telemarketing Success Prediction Using Hyperparameter-Tuned XGBoost
Volume/Issue : 2026;03(06)
Page No : 1138-1147
