Improving Diabetes Forecasting: An Ensemble Approach with Feature Selection in Time Series Analysis

Abstract

Addressing the global health challenge of diabetes through the lens of time series analysis, our study leverages machine learning to advance prediction and management. Introducing various models-linear regression, random forest, gradient boosting, elastic net regression, and support vector regression-implemented in Python with a dataset from GitHub, our methodology emphasizes meticulous data preprocessing and feature selection. Among the explored ensemble techniques, the combination of linear regression, random forest, and gradient boosting stands out, achieving a low mean squared error of 16.17. This underscores the potential of our approach to enhance diabetes prediction accuracy and improve management within time series analysis.