Authors - Ronald S. Cordova, Rowena O. Sibayan, Hazel C. Tagalog Abstract - Digital marketing teams often struggle less with access to algorithms than with choosing the right one for a specific decision. This paper presents a comparative study on the selection of the three most suitable algorithms for two related digital marketing tasks: customer segmentation and promotion-response prediction. Based on the example of Oman's retail industry, a benchmark is established using first-party customer data, including recency, purchase frequency, monetary value, product-category behavior, campaign participation, website visits, and engagement ratio. For customer segmentation, the study focuses on Kmeans, DBSCAN, and Gaussian mixture model because they provide a practical balance of scalability, noise handling, and probabilistic customer-state representation. For promotion-response prediction, the selected models are logistic regression, random forest, and XGBoost because they offer a staged balance between transparency, nonlinear learning, and campaign-ranking performance. For benchmarking and explainability, the same preprocessing approach, leakage prevention, temporal splitting, tuning strategies, and metrics such as silhouette quality, stability, ROC-AUC, PR-AUC, Brier score, calibration, and top-decile lift are employed. Explainability is treated as a condition for adoption rather than an optional reporting activity.
Open Zoom