Optimized LSTM-Based Method for Arrhythmia Detection from ECG Signals Using Crow Search Algorithm}

Abstract

 Accurate detection of cardiac arrhythmias from electrocardiogram (ECG) signals is crucial for the early diagnosis and effective management of cardiovascular disorders. However, the nonstationary, complex, and highly variable nature of ECG signals often limits the performance of conventional classification methods. This study proposes an intelligent arrhythmia detection framework that integrates advanced signal preprocessing, hybrid feature extraction, and an optimized Long Short-Term Memory (LSTM) neural network. In the preprocessing stage, a Butterworth filter is applied to suppress low-frequency noise, while wavelet transformation is used to remove high-frequency interference and enhance the quality of the ECG signals. A comprehensive feature set is then constructed by combining temporal, statistical, and frequency-domain information extracted mainly from the third and fourth levels of wavelet decomposition. To improve classification performance, the hyperparameters of the LSTM network are optimized using the Crow Search Algorithm (CSA), enabling the model to capture complex temporal dependencies more effectively. The proposed framework is evaluated through extensive simulations using confusion matrices, receiver operating characteristic curves, and standard performance measures, including accuracy, precision, recall, and F1-score. The model achieves an outstanding classification accuracy of 99.8\%, indicating excellent robustness, reliability, and practical applicability. The results demonstrate that the integration of effective denoising, multidomain feature extraction, and CSA-optimized LSTM learning provides a powerful and efficient solution for automated arrhythmia detection. This framework has significant potential to support intelligent cardiac monitoring systems, improve clinical decision-making, and enhance the quality of patient care.