Optimizing Marketing Campaigns Using A Hybrid Model Of XGBOOST And Hyperparameter Tuning

Abstract

In digital commerce, achieving the best possible results from marketing campaigns depends on the accuracy of targeting and making decisions based on data. In this study, a hybrid approach of combining XGBoost with multi-strategy hyperparameter optimization (Grid Search, Bayesian Optimization, and Genetic Algorithms) is introduced to improve predictive accuracy for campaign results. This model uses customer behavioral data, an RFM Segmentation, and advanced Ensemble Learning. Two benchmark datasets were used: Bank Marketing and an E-Commerce Customer dataset. Bayesian Optimization outperformed the default XGBoost model with an AUC-ROC of 0.941 and an F1-Score of 0.897 on the Bank Marketing dataset, a 4.2% increase over the default XGBoost model, and a higher score than tuned Logistic Regression (F1-Score 0.821), Random Forest (0.843), and CatBoost (0.862). On the E-Commerce dataset, Bayesian Optimization outperformed the default setting with a 2.2% higher F1-Score of 0.893. The ROI for a simulated campaign targeting the top 20% of customers whose likelihood of responding was predicted by the model was 347%, whereas a campaign targeting at random and a campaign targeting based on baseline Logistic Regression both achieved ROIs of 218% and 291%, respectively. The campaign results, account balance, and customer age were identified as the most important indicators and were emphasized in the Feature Interpretability analysis using the SHAP tool, which enabled actionable insights and campaign strategy adjustments. The results highlight the benefits of combining XGBoost with systematic hyperparameter tuning in predictive marketing analytics and in decision making for scalable, interpretable and economically significant marketing campaigns.