Enhancing Model Accuracy in Diabetes Prediction Through Hybrid Feature Selection

Authors:
Publisher website
Google Scholar
Abstract
Diabetes mellitus remains a significant global health issue, necessitating early and accurate prediction for effective management and reduced complications. Machine learning (ML) models have been applied to predict diabetes, but achieving high accuracy is challenging due to complex and high-dimensional datasets. Effective feature selection is critical to improving both model accuracy and interpretability. This chapter explores a novel hybrid feature selection approach combining filter, wrapper, and embedded techniques to enhance diabetes prediction models. The selected method optimizes feature selection by first removing irrelevant features (filter), selecting optimal subsets (wrapper), and refining features during model training (embedded). We applied this hybrid approach to a diabetes dataset, utilizing patient records with features like age, BMI, blood pressure, and glucose levels. Models such as logistic regression, random forests, and XG Boost were evaluated using metrics like accuracy, precision, recall, F1-score, and AUC-ROC

This publication has 8 references indexed in Scilit: