Machine Learning Techniques for Diabetes Prediction: A Comparative Analysis

Abstract

Diabetes mellitus, characterized by chronic hyperglycemia, presents significant challenges due to its associated complications and increasing morbidity rates. This study examines a range of machine learning algorithms such as Naïve Bayes, Decision Tree, Logistic Regression, Random Forest, Neural Network, Support Vector Machine, LogitBoost, and Voting classifier to develop accurate predictive models for diabetes. The data used in this research is drawn from a comprehensive dataset available on mendeley.com, sourced from the laboratory of Medical City Hospital in Iraq. The focus of the study is on feature selection and evaluation metrics to effectively gauge model performance. Eight classification techniques are employed and compared, including Decision Trees (DT), Random Forests (RF), and LogitBoost. The study's findings highlight DT and RF as the top-performing algorithms, demonstrating comparable predictive abilities, with LogitBoost also showing promising results. Conversely, Support Vector Machine (SVM) shows reduced performance due to its sensitivity to outliers. These insights enable healthcare practitioners to adopt appropriate machine learning methods to improve diabetes prediction, thus enabling timely interventions and enhancing patient outcomes.