Enacting a Hybrid Ensemble Framework to Predict and Prevent Diabetes: Combining ANN, ETC, Random Forest, and Stack-CNN

Worldwide, diabetes mellitus, a lifelong metabolic illness, is an alarming concern for the wellness of society. Complications can potentially be avoided or postponed with early diabetes diagnosis and treatment. Our work recommends a Hybrid Ensemble Framework with a stacking technique that combines the predictions of machine learning and deep learning models to allow for very accurate early-stage disease identification. This framework seeks to achieve very high precision in diabetes prediction. The proposed Hybrid Ensemble Framework model combines predictions from DL and ML, the deep and machine learning base-level architectures, such as Random Forests (RF), Artificial Neural Networks (ANN), and Extra Tree Classifier (ETC), and an ensemble model stacking CNN as the core of the Hybrid Assemblage Framework to improve diabetes predictions trained on a multi-class Iraqi Diabetes Patient Dataset (IDPD-I). Data preprocessing includes handling missing data, Min-Max scaling, SMOTE to correct class imbalance, encoding categorical features and target labels, and removing infrequent classes to enhance data quality. The pertinent features are extracted from the data using the Extra Tree Classifier (ETC) model. The conduct of the suggested Hybrid Ensemble Framework is examined according to key evaluation measures, including F-score, ROC/AUC, accuracy, sensitivity, precision, specificity, and Matthews Correlation Coefficient (MCC). Using the IDPD-I dataset, the CNN's stack demonstrated strong conduct, with an accuracy rating of 98.54%. The findings show that the suggested Hybrid Ensemble Learning Framework performs better than earlier research regarding diabetes forecasts.