TDAPON: AN AI-AYURVEDA HYBRID FRAMEWORK FOR PERSONALIZED DIABETES MANAGEMENT

Abstract

               The Tri-Aware Predictive Optimization Network (TDAPON), an AI-based framework which harmoniously integrates ancient Ayurvedic wisdom with the most current machine learning technologies to offer personalized and adaptive diabetes management care. The model uses Constitution-Aware Feature Encoding based on Ayurvedic ontology and natural language processing to produce patient-specific features: constitution scores, patient-determined (lifestyle) variables, patient-reported symptoms/injury, and patient lab-derived biochemical markers. The rich features comprise the input to the Deep Reinforcement Learning (DRL) module, a decision-making engine that learns about optimal intervention strategies which constantly receive updates from patient participation/input. After the system learns patients in DRL, the resulting interventions take form via the Multi-Objective Genetic Algorithm (MOGA) addressed towards multiple competing health-care related goals: glycemic control, balance of dosha, patient-adherent interventions, and wellness. In MOGA, the TDAPON derives and evolves Pareto-optimal care strategies. Performance assessments indicate that TDAPON greatly exceeds traditional models, such as Decision Tree, Random Forest, and Deep Neural Networks on important measures with accuracy of 96.3%, recall of 95.2%, F-measure of 95.4% and 90% personalization score. There were also positive results using clinical measures, including 21% reduction in HbA1c, 0.82 dosha balance scores, and in 88% of patients. The findings also indicate there is the possibility of marrying the rapidly developing field of AI with ancient healthcare through TDAPON to create a culturally relevant, interpretative, and the human side remained of PA to achieve effective personalized chronic disease management. The framework represents a new area of development for integrative, knowledge-guided, and adaptive healthcare systems.