RETRACTED ARTICLE: Electrocardiogram-based heart disease prediction using hybrid deep feature engineering with sequential deep classifier

Several approaches can detect early heart problems. Electrocardiograms (ECGs) are better and more affordable for early heart disease prediction. ECG data can better predict heart diseases and abnormalities. Standard methods are slow and limited to a narrow ECG signal pattern range. Newer deep learning-based algorithms require tedious and unimportant human feature extraction from raw ECG data. This study presents a new electrocardiogram-based cardiac disease detection system using an ensemble method for feature extraction and deep learning for classification. The approach includes pre-processing of input ECG signals, hybrid feature engineering, and disease categorization. Pre-processing the input signal preserved its integrity by reducing the artifact and noise using a robust filtering technique. ECG feature extraction should use generic automation and beat-specific features. We designed the hybrid feature engineering approach using a Convolutional Neural Network (CNN) and Stationary Wavelet Transform (SWT). The SWT method employs dynamic thresholding to segment ECG data into individual heartbeats, from which the beat-specific feature vector is derived. To automate the process of classifying cardiac diseases, a powerful 3-layer CNN has been trained to do so and now extracts those attributes automatically. Manifold learning of feature selection is then used to normalize the combined features from the two methods. Long Term Short Memory (LSTM), a deep learning classifier, has been given the ensemble feature vector. For LSTM, the sequential characteristics vector transmits information to the LSTM layer and then calculates the probabilities for each class to provide precise disease classification in the classification phase. The suggested model surpasses cutting-edge deep learning-driven techniques when used with a publicly available multi-disease ECG dataset.