Predicting Customer Purchase Using Machine Learning: A Comparative Study of Traditional and Deep Learning Models

This study provides a comprehensive evaluation of the effectiveness of both traditional machine learning techniques and deep learning models in predicting customer purchase behavior within a data-driven business environment. Using a simulated dataset comprising 5,000 customers, the study incorporates a range of demographic variables (such as age, gender, and income level) alongside behavioral features, including browsing frequency, purchase history, and engagement patterns. Six predictive models are systematically examined: Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), a feed-forward Neural Network, and a Long Short-Term Memory (LSTM) network. Model performance is assessed using multiple evaluation metrics accuracy, F1-score, receiver operating characteristic area under the curve (ROC-AUC), and training time to ensure a balanced comparison of predictive power, robustness, and computational efficiency. The results indicate that deep learning models consistently outperform traditional machine learning approaches in terms of predictive accuracy and overall classification performance, with the LSTM model achieving the highest scores due to its ability to capture temporal and sequential patterns in customer behavior. Nevertheless, traditional models demonstrate notable advantages, particularly in faster training times, lower computational cost, and greater interpretability, making them suitable for organizations with limited resources or strong explainability requirements. Overall, the findings offer practical insights for businesses and data practitioners by highlighting the trade-offs between accuracy, interpretability, and computational efficiency, thereby supporting informed decision-making when selecting predictive analytics models for customer behavior forecasting.