AI-Powered Diagnostics: Unveiling Machine Learning Models for Diabetes Prognosis

Abstract

Diabetes is a chronic metabolic disorder characterized by elevated blood glucose levels due to insufficient or ineffective insulin secretion. This disruption affects the body's ability to process carbohydrates, fats, and proteins, leading to severe complications if left undiagnosed or untreated. Early detection of diabetes is essential for minimizing associated health risks and reducing its overall prevalence. With advancements in artificial intelligence and machine learning, predictive models have become increasingly effective in diagnosing diabetes at an early stage.This research explores various machine learning approaches for early-stage diabetes prediction, focusing on feature selection, dimensionality reduction, and multiple classification techniques. A relief-based filter method (ReliefF) is utilized for feature selection, which enhances model performance by identifying and prioritizing significant attributes. Among the predictive models applied, Random Forest (RF) emerges as the most accurate classifier, achieving a precision of 98.5%. This highlights its superior capability in distinguishing diabetic and non-diabetic cases. Support Vector Machine (SVM) follows closely with a precision of 96.6%, while Neural Network (NN) achieves 96.2%, effectively capturing intricate data patterns.To ensure a robust performance assessment, the models undergo evaluation using tenfold cross-validation, accuracy scores, confusion matrices, and classification reports. The study’s findings confirm that Random Forest outperforms SVM and NN in terms of diagnostic accuracy, making it a highly effective tool for early diabetes detection. These results emphasize the importance of machine learning in enhancing predictive healthcare, potentially leading to improved patient outcomes and timely medical interventions.