High Performance ECG Signal Analysis via Advanced ML Techniques

This study presents a hybrid HMM–Boosting frame-work for accurate classification of electrocardiogram (ECG) signals using data from the MIT-BIH Arrhythmia Database. The proposed model integrates wavelet-based denoising, Autocorrelation (AC) and Partial Autocorrelation Function (PACF) feature extraction, and Hidden Markov Models (HMMs) for temporal sequence modeling. HMM-derived features are combined with morphological descriptors and processed through ensemble classifiers (AdaBoost and Gradient Boosting) to enhance diagnostic precision. Dataset partitioning was performed on a patient-wise basis to prevent data leakage, while GroupKFold cross-validation ensured robust and unbiased evaluation. The hybrid model achieved an accuracy of 99.7%, outperforming the baseline HMM. Statistical tests and fold-level t-tests confirmed that the improvements were highly significant. Error analysis revealed minor confusion between Normal and Bradycardia classes, indicating areas for further improvement. Overall, the findings demonstrate that combining HMM-based temporal modeling with Boosting ensembles substantially enhances arrhythmia detection accuracy and clinical reliability in automated ECG analysis.