A Machine Learning Approach for A Customized Product Recommendation System in E-Commerce

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The rapid growth of e-commerce platforms has intensified competition and increased the need for personalized product recommendation systems that enhance user experience and engagement. This study aims to design and develop a machine learning–based personalized recommendation system by analyzing user behavior and product data, examining product related features, and implementing an effective predictive model. Traditional recommendation systems, such as content-based and collaborative filtering, often suffer from limitations including cold-start problems, data sparsity, and lack of contextual awareness, particularly in developing economies like Nigeria. To address these challenges, this study adopts a data driven approach using a dataset of over 10,000 products scraped from a Nigerian e-commerce platform, alongside 10,000 simulated user interactions from 800 users. The dataset captures diverse product attributes, pricing ranges (₦3,500 to ₦7,000,000), and contextual information relevant to user preferences. Through exploratory data analysis, patterns in user behavior and product characteristics were identified to support personalized recommendation. A machine learning model was designed and implemented using Light Gradient Boosting Machine (LightGBM) and compared with a Deep Neural Network (DNN) based on Neural Collaborative Filtering. The models were evaluated using standard performance metrics including precision, recall, and F1-score. The results indicate that LightGBM significantly outperformed the DNN, achieving an F1@10 score of 0.87 compared to 0.66, demonstrating its effectiveness in capturing user preferences through engineered features such as price sensitivity, location, and product popularity. The study concludes that feature-based machine learning models provide more accurate and efficient recommendations in data-constrained environments than deep learning approaches. A web-based prototype developed using Flask further demonstrates the practical applicability of the system, enabling real-time personalized recommendations. This research contributes a scalable and context-aware recommendation framework suitable for e commerce platforms in emerging markets.