Hybrid GNN and Random Forest-based Drug Recommendation System using Machine Learning Techniques for Proactive Side Effect Mitigation

Abstract

Adverse drug reactions and ineffective medication selection remain major challenges in modern healthcare systems, especially for patients with multiple medical conditions and varying physiological characteristics. Existing drug recommendation systems primarily focus on treatment effectiveness while giving limited importance to patient safety and side effect mitigation. To address this issue, this research proposes a hybrid intelligent drug recommendation framework titled “GNN-based Drug Recommendation using Machine Learning Techniques for Proactive Side Effect Mitigation.” The proposed system integrates Graph Neural Networks (GNNs) and Random Forest-based Machine Learning techniques to recommend safer and more effective medications by analyzing complex relationships among diseases, drugs, patient history, and side effects. The system accepts patient-specific inputs such as age, gender, disease name, and pre-medical conditions, and dynamically predicts the most suitable medication accordingly. Unlike traditional recommendation systems, the proposed model actively evaluates both treatment effectiveness and potential adverse effects before generating recommendations. A heterogeneous medical graph is constructed to model interactions between diseases, drugs, and side effects, enabling the GNN to learn hidden relational patterns from interconnected medical data. The extracted graph embeddings and patient features are further processed using the Random Forest algorithm to predict allopathy drug names, generic drug names, side effect names, side effect scores, and effectiveness scores. The recommendation changes adaptively based on patient demographics and medical history, thereby supporting personalized medicine. The proposed approach aims to improve patient safety, reduce adverse drug reactions, and enhance the overall reliability of intelligent healthcare recommendation systems. Experimental analysis demonstrates that combining relational learning through GNNs with predictive capabilities of Random Forest provides more context-aware and patient-centric drug recommendations compared to conventional machine learning approaches. The proposed system also predicts suitable generic medicine alternatives along with branded allopathy drugs, thereby supporting cost optimization and improving medicine affordability for patients without compromising treatment effectiveness and safety.