Electrocardiogram (ECG)-based biometric authentication systems offer intrinsic resistance to spoofing due to their physiological uniqueness. However, their performance in dynamic real-world settings, such as wearable devices or stress-induced conditions, is often compromised by noise, electrode displacement, and intra-subject variability. This study proposes a novel hybrid framework that enhances robustness, ensuring high authentication accuracy and reliability in adverse conditions, through integrated wavelet-based signal processing for noise suppression and a deep-learning classifier for adaptive feature recognition. The system employs preprocessing, QRS complex detection, distance–deviation modeling, a statistical comparison method that quantifies morphological similarity between ECG templates by analyzing amplitude and shape deviations and an averaging-threshold mechanism, combined with a feedforward Multi-Layer Perceptron (MLP) neural network for classification. The MLP is trained on extracted ECG features to capture complex nonlinear relationships between waveform morphology and user identity, ensuring adaptability to variable signal conditions. Experimental validation on the ECG-ID dataset achieved 98.8% accuracy, 95% sensitivity, an Area Under the Curve (AUC) of 0.98, and a low false acceptance rate, outperforming typical wearable ECG authentication systems that report accuracies between 90% and 95%. With an average processing time of 8 seconds, the proposed method supports near real-time biometric verification suitable for healthcare information systems, telehealth platforms, and IoT-based access control. These findings establish a scalable, adaptive, and noise-resilient foundation for next-generation physiological biometric authentication in real-world environments

Keywords: electrocardiogram biometrics, wavelet decomposition, QRS complex detection, feedforward neural network, deep learning classification, noise-resilient authentication, biometric security