Intelligent Symptom-to-Medicine Mapping using MAGRU-IDMO: An Attention-GRU Framework with Bio-Inspired Optimization

Personalized symptom-to-drug recommendation systems match medications to an individual’s symptoms, medical history, and genetic profile. Using deep learning and knowledge graphs, they improve precision medicine, reduce trial-and-error prescribing, and minimize adverse drug reactions. This study proposes MAGRU-IDMO, a hybrid framework combining multi-head attention with gated recurrent units (GRUs), optimized via the improved dwarf mongoose optimization (IDMO) algorithm. The model is trained on two healthcare datasets—medical recommendations and drug data including uses, side effects, and reviews— preprocessed with cleansing, stemming, and stopword removal using the Sastrawi NLP library. Multi-head attention captures diverse sequence dependencies, GRUs model temporal correlations, and IDMO refines feature selection and optimization by simulating mongoose behavioral dynamics. Experiments show MAGRU-IDMO outperforms LSTM and attention-only models in accuracy and efficiency, offering strong potential for adaptive, interpretable medical decision-making. Future work will incorporate real-time patient feedback and clinical validation.