An effective hybrid optimal deep learning approach using BI-LSTM and restricted Boltzmann machines whale optimization to detect arrhythmia

Electrocardiography (ECG) is a widely recognized noninvasive method employed in the field of medical science to gather data pertaining to the cardiac rhythm and the present state of the heart. The utilization of automated ECG arrhythmia diagnosis has proven to be beneficial in reducing the workload of clinicians and enhancing the efficacy and efficiency of diagnoses. In line with this objective, the present study introduces a hybrid optimal deep learning (DL) approach for the purpose of addressing the recognition of arrhythmias. To improve the results of long ECG (electrocardiogram) data classification, a sequential pre-processing approach has been developed that uses UBF (unified bilateral filtering), Z-normalization, and FCM (fuzzy c-means)-based segmentation to construct the synthetic signal. Then, a synthetic signal based on a GAN (generative adversarial network) is created to manage an unbalanced signal class. An original hybrid strategy has been suggested for effectively detecting arrhythmia to use Bi-LSTM (Bi-Directional Long Short-Term Memory) and RNN-rBM (restricted Boltzmann machines with recurrent neural network). The hyper-parameters of hybrid Bi-LSTM- RNN-rBM have been optimized using WO (Whale optimization) that improves the accuracy of classification results and proposed work is named as WO-Bi-LSTM-RNN-rBM. This technique enhances categorization performance with less complexity by including even more intricate and distant aspects of the ECG signal at each convolution operation without increasing network parameters. The experimental results demonstrate that the suggested RNN-rBM model surpasses current models with 99.5% accuracy, 98.2% F1, 98.2% exactness, and 98.5% recall in learning for MIT-BIH, given ECG data to diagnose arrhythmia. Overall, the WO-Bi-LSTM-RNN-rBM offers a high-performance automated detection strategy to detect arrhythmia and an inexpensive ECG signal-reducing method.