A Comparison Study of Machine Learning and Deep Learning Approaches for Diabetes Mellitus Prediction

Diabetes is a chronic disease, a collection of metabolic illnesses due to elevated blood sugar levels. If an accurate early prediction is feasible, the risk and severity of diabetes can be significantly decreased. Due to the insignificant labeled data and outliers in the diabetes datasets, developing a robust and reliable diabetes prediction model is challenging. This paper develops a robust hybrid framework for diabetes prediction using machine learning and deep learning approaches. Several ML and DL models are ensembled to enhance diabetes prediction. For the experimental analysis of the work, the Pima Indian Diabetes Dataset (PIDD) is used. The proposed ensembling classifier outperforms the existing models by 2.00% in AUC. It is the best-performing classifier from all the extensive experiments, with sensitivity, specificity, F1 score, false positive rate, and AUC of 0.789, 0.934, 0.092, 66.234, and 0.950, respectively. Among all the models, DL outperforms the other models for early diabetes prediction. As a result, the proposed ensembled models provide a useful predictive tool for medical assistance. The findings may be utilized to develop an automated predictive tool that can aid in the early diagnosis of the illness.