International Journal of Advanced Research in Computer and Communication Engineering

Abstract: In this paper, we present an Intelligent Medicine Recommendation System with Salt Composition Analysis (IMRS-SCA), production-ready with FastAPI-based deployment of multi-label classification combined with semantic matching. We create a consolidated dataset of 253,973 Indian pharmaceutical products with compositional metadata along with 14,683 disease-drug associations aggregated from national pharmaceutical databases and clinical prescription records. Our system uses a three-tier matching pipeline extending the raw pharmaceutical attributes comprising salt composition, manufacturer, and disease indications through a normalization framework that incorporates 28+ canonical salt forms, fuzzy string matching with a threshold ≥ 0.85, and synonym-aware semantic encoding. Training was done using a Random Forest multi-output classifier with stratified train-validation-test splits in order to handle class imbalance across more than 100 disease categories. The proposed model gives F1-Score = 0.9108, Precision = 0.9269, Recall = 0.8996, and a mean confidence score = 0.91 for top-ranked recommendations on the reserved test set, outperforming baseline exact-match retrieval (Precision = 0.52). FastAPI-based deployment achieved a mean response latency of 120 ms per query under concurrent load, confirming suitability for real-time clinical decision support. In ablation studies, maximum marginal gain was observed due to the salt normalization layer, which improved alternative medicine discovery for generic substitution scenarios by 34%. The system requires no proprietary medical data, runs on commodity hardware with a

Downloads: | DOI: 10.17148/IJARCCE.2025.141114