ARTIFICIAL INTELLIGENCE–DRIVEN MULTIMODAL FRAMEWORK FOR DRUG RECOMMENDATION AND DELIVERY PREDICTION USING CLINICAL AND DRUG INFORMATION

Abstract

Drug recommendation and healthcare decision support require the integration of heterogeneous data sources including drug characteristics, clinical feedback, patient reviews, and therapeutic information. Conventional prediction approaches frequently rely on single-source information and may not adequately capture complex relationships influencing recommendation outcomes. This study proposes a multimodal machine learning framework for drug recommendation prediction through the integration of clinical reviews and drug-related information. Multiple publicly available datasets containing drug composition, therapeutic uses, side effects, manufacturer information, clinical reviews, and patient feedback were collected and fused into a unified multimodal dataset. The preprocessing pipeline included data cleaning, missing value handling, feature normalization, drug name standardization, multimodal fusion, and duplicate removal. Textual information was transformed using Term Frequency–Inverse Document Frequency (TF–IDF) and transformer-based semantic embeddings, while numerical clinical attributes were encoded and integrated using feature fusion techniques. Dimensionality reduction through Singular Value Decomposition (SVD) was additionally applied to improve representation efficiency. Several machine learning and deep learning approaches were evaluated, including Random Forest, XGBoost, SVD-enhanced XGBoost, weighted binary classification, and BERT-based multimodal classification. Experimental evaluation was performed using accuracy, precision, recall, and F1-score. Comparative analysis demonstrated that transformer-enhanced multimodal learning improved predictive performance, achieving a maximum classification accuracy of 68.14%, while the SVD-enhanced framework achieved the strongest balanced performance with an F1-score of 54.95%. Weighted learning further improved minority-class detection capability. The findings indicate that multimodal integration of clinical and drug information can support intelligent healthcare analytics and provide a foundation for future AI-assisted drug recommendation and decision-support systems.