Real-Time Myocardial Infarction Detection Using PPG Signals and Machine Learning on Wearable Devices

Abstract

Cardiovascular disease remains the leading cause of death globally, with myocardial infarction (MI) representing its most acute and time-critical manifestation. Current clinical diagnosis relies on hospital-based tools, namely the 12-lead electrocardiogram and serum troponin assays, which are effective but inaccessible during the critical prehospital period when early detection would have the greatest impact on patient outcomes. This thesis investigates whether photoplethysmography (PPG) signals acquired from wrist-worn wearable devices contain sufficient discriminative information to support machine learning-based MI classification, and which classifier is best suited for real-time deployment on resourceconstrained hardware. A publicly available PPG dataset from the UCI Machine Learning Repository was used, containing 2,576 samples with 2,000 amplitude features each, labelled as either Normal or Myocardial Infarction with near-perfect class balance. The experimental pipeline applied StandardScaler normalization, Principal Component Analysis dimensionality reduction from 2,000 to 50 components, and 5-fold stratified cross-validation. Four supervised classification algorithms were evaluated: Logistic Regression, Random Forest, Gradient Boosting, and Support Vector Machine. All four classifiers exceeded 95 percent accuracy on the held-out test set. Gradient Boosting achieved the highest Area Under the ROC Curve at 0.954, confirming its superior ability to capture nonlinear patterns in the PCA-reduced feature space. Logistic Regression achieved an identical accuracy of 95.5 percent with an AUC of 0.946 and is recommended for wearable deployment due to its minimal computational requirements and interpretable decision boundary. These results establish a viable proof-of-concept for continuous, passive MI detection on consumer wearables, with real-world clinical validation identified as the essential next step.