A Machine Learning and Internet of Medical Things Framework for ECG-Based Heart Disease Diagnosis

Abstract

Cardiac arrhythmia is a major global health concern, and conventional electrode-based Electrocardiogram (ECG) monitoring cannot continuously track drivers, limiting early detection of heart-related risks that could cause road accidents. This study presents a Cardiac Health Assistance System (CHAS) combining Internet of Medical Things (IoMT) sensors and Machine Learning for continuous, real-time cardiac monitoring and diagnosis of vehicle drivers. Wearable ECG and physiological sensors capture heart rate, ECG signals, and driver demographics (age, sex, height, weight), wirelessly transmitted via a multi-layer IoT architecture to a cloud platform. After preprocessing, feature extraction, and labelling, a two-stage classification pipeline evaluates combinations of Naive Bayes, weighted K-Nearest Neighbors (KNN), Decision Tree, and Support Vector Machine classifiers, benchmarked against Long Short-Term Memory (LSTM) and a Fuzzy Inference System-LSTM (FLSTM) model using five-fold cross-validation on 9,200 samples. The best two-stage configuration, Naive Bayes followed by SVM, achieved an F1-score of 94.8 percent and 92.5 percent accuracy at the first stage. Across varying training-data proportions, CHAS consistently outperformed LSTM and FLSTM, reaching an overall precision of 96.30 percent, accuracy of 97.35 percent, and F1-score of 95.77 percent. CHAS showed superior and more stable performance than existing LSTM and FLSTM approaches across all evaluated metrics and data sizes, confirming its ability to reliably detect cardiac abnormalities while minimizing false positives and negatives, supporting timely driver alerts. The proposed IoMT and Machine Learning-based CHAS offers an effective, real-time solution for continuous cardiac monitoring of vehicle drivers, improving road safety through early detection and immediate intervention for cardiac disorders.