An Enhanced Deep Machine Learning Framework for Chronic Disease Prediction in IoMT-Powered Healthcare 5.0: A Comprehensive Case Study on Alzheimer’s Disease

Abstract

Chronic diseases such as Alzheimer’s disease remain a critical concern in healthcare, demanding precise and timely prediction systems to support clinical decision-making in IoMT-enabled Healthcare 5.0 environments. In this study, an Alzheimer’s dataset was employed to design a dual-branch framework integrating classification and detection models for enhanced prediction. For the classification branch, deep and machine learning models including ResNet101, DCNN, Xception, NasNetMobile, Ensemble Model, KNN, Random Forest, Logistic Regression, and XGBoost were evaluated. For the detection branch, the YOLO family {YOLOv5, YOLOv8, YOLOv9, YOLOv11} was utilized to localize abnormalities in medical images. Experimental evaluation demonstrated that the Ensemble Model and NasNetMobile achieved superior classification performance, with an accuracy, precision, recall, and F1-score of 99.4%, outperforming traditional classifiers. In detection, YOLOv8 yielded the best results, attaining a precision of 87.3%, recall of 88.1%, and mAP of 91.3%, highlighting its reliability in abnormality localization. Furthermore, Explainable AI (XAI) techniques such as Grad-CAM were applied to generate heatmaps, emphasizing critical image regions contributing to model predictions, thereby improving interpretability and trustworthiness. The integration of classification, detection, and explainability underscores the potential of deep machine learning approaches to advance intelligent, accurate, and transparent Alzheimer’s disease prediction within modern IoMT-driven healthcare systems.