Predicting Diabetes Mellitus in Healthcare: A Comparative Analysis of Machine Learning Algorithms

Abstract

Hyperglycemia is the underlying cause of diabetes, a long-term health condition. This condition may be diagnosed using a battery of chemical and physical testing. Conversely, the eyes, heart, kidneys, and nerves may all suffer damage or even death from undetected and untreated diabetes. Accordingly, the mortality rate may be decreased by the analysis and early identification of diabetes. The efficacy of using ML and DL models for early illness identification has been discovered in various medical domains lately. This study explores a prediction of Diabetes Mellitus in healthcare using the Pima Indians Diabetes dataset (PIDD), comprising 768 instances and 9 attributes. o address class imbalance, the ADASYN technique generates synthetic data for minority classes. F1-Score, precision, recall, and accuracy are the metrics used to train and assess ML models like LR, RF, and KNN. Outcomes show that LR outperforms RF and KNN, achieving the highest accuracy (92.26%), precision (82%), recall (91%), and F1-Score (86%), demonstrating its effectiveness in diabetes prediction and highlighting its potential in improving healthcare decision-making. Future work in diabetes prediction can focus on several key areas to further enhance model performance and applicability in healthcare.