A stacking ensemble approach for diabetes prediction

Abstract

Diabetes Mellitus (DM) is characterized as a metabolic disorder that increases blood glucose levels for longer periods of time. It is vital to continuously monitor these glucose levels; otherwise, they can lead to different types of abnormalities like retinopathy, nephropathy, and neuropathy. Therefore, cutting-edge data analytics and machine learning (ML) approaches are essential for the identification, diagnosis, and treatment of DM. In this paper, a thorough study of diabetes prediction approaches is discussed. Lifestyle, medical history, and genetic history are the factors due to which an individual acquires diabetes. This research aims to develop predictive models for early intervention in diabetes management. A stacking ensemble method is built to improve the classification of diabetes using the PIMA dataset, a well-known biomedical dataset. The stacking ensemble approach combines multiple diverse base approaches, such as logistic regression (LR), support vector machines (SVM), and decision trees (DT), to leverage their collective predictive power. Through comprehensive evaluations, we determine that the stacking ensemble method is superior to individual models and other ensemble techniques in terms of performance metrics. An accuracy of 95.291%, precision of 91.6%, recall of 91.3% and F1 score of 83.2% were obtained. Our findings highlight the potential of the stacking ensemble method as a valuable model for accurate and reliable diabetes classification in biomedical data analysis.