Machine Learning for Early Diabetes Screening: A Comparative Study of Algorithmic Approaches

Abstract

Diabetes mellitus, a chronic metabolic disorder, poses a significant global health challenge. Early screening and risk assessment are crucial for effective management and prevention. This study evaluates the performance of various machine learning models – Artificial Neural Networks (ANNs), Random Forest (RF), k-nearest Neighbors (k-NN), and Support Vector Machine (SVM) – in screening diabetes risk using a dataset based on patient-reported symptoms such as age, gender, polyuria, polydipsia, and sudden weight loss. The dataset, comprising self-reported data from 520 individuals, highlights the potential association of specific symptoms and demographics with diabetes risk. Rigorous analysis demonstrates the superior performance of the RF model in terms of accuracy and F1 Score. Feature importance analysis further emphasizes the critical role of patient-reported symptoms in assessing predisposition to diabetes. The findings suggest that with its robust predictive capability, RF is particularly suitable for early screening, offering valuable insights into symptom-based diabetes risk assessment. This research advances non-invasive, symptom-based screening tools, paving the way for early interventions and tailored prevention strategies.