ADR-W: A Novel Accuracy-Driven Weighting Approach for Accurate and Reliable Diabetic Diagnosis

Abstract

Metabolic diseases manifest as hyperglycemia. It occurs when the corpse's humulin production is insufficient. Glycemia may be fatal if not treated appropriately and detected on time since it threatens the eyes, kidneys, nerves, heart, and blood arteries, among other important bodily organs. Research in computational diabetes has shown that machine learning can accurately predict who will get diabetes. Nevertheless, the current accuracy rate indicates that there is enough opportunity for enhancement. Using the three datasets provided, develop a machine learning system capable of diabetes prediction and diagnosis. A sequential strategy to improving predictive modeling's categorization accuracy. In the first phase, we employ preprocessing techniques such as Hot Deck Imputation (HDeckImp) to efficiently deal with missing variables and reduce classification errors. In the second stage, K-fold cross-validation is employed to ensure that the model is durable and adaptable to new scenarios. The third stage uses four traditional machine learning models to make correct predictions. In the final stage, Accuracy-Driven Reinforced Weighting (ADR-W) is applied to increase ensemble performance depending on expected accuracy. The suggested methodology aims to improve the accuracy and dependability of predictions in classification tasks. With its 98.5% accuracy on the Frankfurt dataset, the proposed model proved to be valuable. ADR-W's model to reduce processing time further enhances the system's functionality.