Prediction of Blood Sugar Levels in Type 2 Diabetes Mellitus Patients Based on Diet and Medication Compliance Using Naive Bayes and BAT Algorithms

Abstract

Type 2 diabetes mellitus poses a significant global health especially in Indonesia  challenge, primarily due to patient non-adherence and limited monitoring. Therefore,  technology-based approaches play a crucial role in detecting potential blood sugar  elevations early, enabling faster and more targeted interventions. This study introduces an integrated predictive framework that combines a Naive Bayes classification algorithm with a Bat-inspired metaheuristic (BAT) for automated feature selection. Optimized by the BAT algorithm, the system achieved high performance: 95% accuracy, 0.94 precision, 0.96 recall, 0.95 F1 score, and 0.90 Cohen's Kappa, indicating near-perfect agreement with actual outcomes. These results confirm the potential of the Naive Bayes and BAT approaches as reliable clinical decision support tools for proactive diabetes management.