An enhanced hybrid GCNN-MHA-GRU approach for symptom-to-medicine recommendation by utilizing textual analysis of customer reviews

Medication recommendation is essential in improving patient treatment and minimizing the occurrence of undesirable effects, yet current approaches prove to be incompetent in addressing sophisticated relationships between syndromes and customer feedback. This study mitigates this gap by presenting a sophisticated model of symptom-to-medicine drug suggestion that deploys state-of-the-art machine learning algorithms for enhanced precision and customization in offering drug suggestions. The novelty lies in the combination of customer reviews with a hybrid model made up of Graph Convolutional Neural Networks (GCNNs), multi-head attention, and Gated Recurrent Units (GRUs) to extract complex relationships and sequential dependencies. The Competitive Game Optimizer also further optimizes recommendations to provide solid and personalized treatment recommendations. The approach includes text preprocessing, numerical transformation via TF-IDF and Word2Vec, and evaluation against baseline models using accuracy, precision, recall, and F1-score. Key results show the better performance of the model with 95.65% accuracy, F1 score of 95.12%, and PRAUC of 0.9857, reflecting outstanding precision-recall trade-offs. The Jaccard similarity index of 0.9514 and mean average precision of 0.9725 reflect the effectiveness of the model in providing relevant recommendations. The results highlight the importance of the combination of varied data sources and sophisticated optimization methods, enabling better patient outcomes and revolutionary possibilities in healthcare systems.