A Hybrid ECG-PPG Fusion Model for Cardiac Arrhythmia Detection

Abstract

Cardiac arrthymia is a major healthcare concern that requires continuous monitoring and timely diagnosis. Multi-modal physiological signal analysis has emerged as a promising approach to enhance diagnostic reliability by leveraging complementary information from electrocardiogram (ECG) and photoplethysmogram (PPG) signals. This study presents a hybrid framework for early arrhythmia prediction through ECG–PPG signal fusion integrated with machine learning and deep learning techniques. Raw signals are acquired in WFDB format and undergo preprocessing, including synchronization, normalization, and fixed-length standardization. Statistical feature selection using ANOVAbased SelectKBest is employed to extract discriminative features, followed by multi-modal feature fusion to construct an optimized input space. The classification stage utilizes Gaussian Naïve Bayes, XGBoost, Deep Neural Networks (DNN), and RNN-LSTM models, while regression analysis incorporates Bayesian Ridge, Random Forest Regressor, and XGBoost Regressor for heart rate estimation. Data imbalance is addressed using SMOTE, and performance is evaluated using standard classification and regression metrics. Experimental results demonstrate that the XGBoost classifier achieves superior performance with 99.2% accuracy and 99.8% AUC, while the XGBoost regressor attains an R² score of 96.6% with minimal error rates. A Flask-based web application is developed to enable real-time inference, automated preprocessing, and clinical report generation through a secure user interface.