An Innovative Ensemble Deep Learning Clinical Decision Support System for Diabetes Prediction

Abstract

Diabetes is a significant global health concern, with an increasing number of diabetic people at risk. It is considered a chronic disease and leads to a significant number of fatalities annually. Early prediction of diabetes is essential for preventing its progression and reducing the risk of severe complications such as kidney and heart diseases. This study proposes an innovative Ensemble Deep Learning (EDL) clinical decision support system for diabetes prediction with high accuracy. The proposed EDL model uses Deep Learning (DL) architectures such as Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), and Convolutional Neural Networks (CNN), integrated with an ensemble learning-based stacking model. The EDL is implemented based on a stack ensemble model that applies meta-level models, including stack-ANN, stack-CNN, and stack-LSTM, to improve the prediction of diabetes. Three diabetes datasets, such as I. Pima Indian Diabetes Dataset (PIMA-IDD-I), II. Diabetes Dataset Frankfurt Hospital Germany (DDFH-G), and III. Iraqi Diabetes Patient Dataset (IDPD-I) are used to train the novel EDL models. The Extra Tree Classifier (ETC) approach is used to extract the relevant features from the data. The performance of the proposed EDL models is evaluated based on major evaluation metrics such as accuracy, precision, sensitivity, specificity, F-score, Matthews Correlation Coefficient (MCC), and ROC/AUC. Among the proposed EDL models, the stack-ANN achieved robust performance using DDFH-G, PIMA-IDD-I, and IDPD-I datasets with accuracy scores of 99.51%, 98.81%, and 98.45%, respectively. The overall results demonstrate that the proposed EDL models outperform previous studies in predicting diabetes.

Keywords

• Artificial neural networks
• convolutional neural networks
• diabetes mellitus
• deep learning
• ensemble learning
• long short-term memory