Boosting-based machine learning approaches for diabetes prediction using Indian demographic and health survey-2021 data

Abstract This study examines the effectiveness of boosting-based machine learning classifiers in identifying diabetes in the Indian population. Traditional diagnostic methods for diabetes are time-consuming and prone to errors, and computer-aided diagnostic techniques can improve performance and reduce costs. The study analyzes data from the Indian Demographic and Health Survey 2021, focusing on women aged 19–49 who tested positive for diabetes. The dataset includes clinical, anthropometric, and biochemical components, and 12,103 positive responders was considered. The study adopts predictive exploration-based boosting machine learning models, including adaptive boosting, categorical boosting, extreme gradient boosting, gradient boosting, and light gradient boosting models. Feature extraction is performed using kernel principal component analysis. The extreme gradient boosting model performs well on the dataset, with accuracy, f1-score, precision, and recall values of 81% and 83%, 81% and 82%, 81% and 88%, and 81% and 76%, respectively, before and after applying kernel principal component analysis. Adaptive boosting performs poorly, while categorical, gradient, and light gradient boosting models perform moderately. The study achieves maximum accuracy, precision, the area under the curve, and recall.