Predicting Pharmaceutical Product Discontinuation Using Decision Tree and Random Forest Algorithms Based on Product Attributes

Abstrak

This study aims to predict the discontinuation of pharmaceutical products using machine learning models, focusing on key product attributes such as manufacturer, composition, price, and packaging. A comprehensive dataset of over 250,000 pharmaceutical products from India was analyzed, with two models—Decision Tree and Random Forest—being employed for prediction. The models were evaluated based on accuracy, precision, recall, and F1-score. The Random Forest model outperformed the Decision Tree with a higher accuracy, but both models struggled with the imbalanced dataset, showing low recall for the minority class (discontinued products). Feature importance analysis identified manufacturer and composition as the most influential factors in predicting product discontinuation. These findings offer valuable insights for pharmaceutical companies in managing product portfolios and optimizing their lifecycle strategies. Despite limitations in data quality and class imbalance, this study provides a foundation for future research, suggesting the integration of additional data sources and the application of deep learning techniques to further enhance prediction accuracy.