A Hybrid Machine Learning Framework for Marketing Analytics and Campaign Optimization: Integrating Prediction, Interpretation, and Behavioral Pattern Mining

In the era of data-driven marketing, optimizing customer engagement and campaign effectiveness remains a critical challenge. This study proposes a novel hybrid machine learning framework designed to enhance marketing analytics through three integrated stages: accurate churn prediction, model interpretability via SHAP values, and behavioral pattern mining using association rules. Utilizing a dataset of consumer interactions, we compare the performance of logistic regression, random forest, XGBoost, and a custom ensemble model, achieving a maximum ROC-AUC of ۰.۸۵۵ with the hybrid approach. Our framework not only improves predictive accuracy but also enables actionable insights by identifying key drivers of customer churn and uncovering cross-selling and up-selling opportunities through frequent itemset mining. Compared to a baseline Q۱ study (Shi et al., ۲۰۲۳), our methodology offers greater generalizability, transparency, and practical applicability across industries. The results highlight the value of combining predictive modeling with explainable AI and market basket analysis to inform personalized and dynamic marketing strategies. This research contributes both theoretically and managerially by bridging the gap between advanced analytics and real-world marketing execution.

Keywords:

Customer churn prediction , Hybrid machine learning , SHAP values , Association rule mining , Personalized marketing