Direct marketing campaigns remain an essential strategy for financial institutions aiming to increase customer engagement and product adoption. Traditional modeling approaches for predicting customer conversion often rely on linear assumptions and limited feature interactions, which may not fully capture the complex behavioral patterns observed in real-world marketing data. With the growth of data-driven decision-making, machine learning methods provide new opportunities to improve the accuracy and efficiency of customer targeting strategies.This thesis evaluates the performance of five machine learning models—Logistic Regression, Decision Tree, K-Nearest Neighbors, Multilayer Perceptron, and XGBoost—in predicting customer subscription to a term deposit product using the Bank Marketing dataset. The study conducts comprehensive exploration data analysis, implements standardized preprocessing and feature engineering procedures, and examines multiple evaluation metrics including accuracy, precision, recall, F1-score, and ROC-AUC.Results indicate that ensemble-based methods, particularly XGBoost, outperform traditional linear and distance-based models in capturing complex relationships among customer demographics, financial attributes, and campaign-level features. Logistic Regression provides interpretable baseline performance, while neural network and tree-based methods achieve competitive predictive accuracy. The findings highlight the potential for advanced machine learning approaches to enhance direct marketing effectiveness by improving customer targeting and reducing unnecessary contact costs.This work contributes to the broader literature on applied machine learning in marketing analytics and demonstrates the practical value of predictive modeling in optimizing campaign outcomes. The thesis also discusses ethical considerations, modeling limitations, and future directions for expanding algorithmic decision-making in customer engagement strategies.