Advanced Heart Disease Prediction Through Spatial and Temporal Feature Learning with SCN-Deep BiLSTM

Heart disease prediction using machine learning methods faces various challenges, such as low data quality, missing irrelevant values, and underfit and overfit problems, which increase the time complexity and degrade the model's prediction performance. Moreover, the hybrid models for heart disease prediction showed poor accuracy due to the irrelevancy in the dataset. Therefore, a search optimizer with a deep convolutional neural network coupled with a Deep Bidirectional long short-term memory classifier (SCN-Deep BiLSTM) is proposed to handle the abovementioned issue. The importance of SCN-Deep BiLSTM relies upon establishing the spatial information and temporal features from the ECG signals that support learning while minimizing the computational complexity associated with learning from raw signals.The SCN-Deep BiLSTM model achieves the accuracy, F-score, precision, recall, and critical success index of 0.97, 0.97, 0.98, 0.99, and 0.97, respectively for 80% of model training, whereas the SCN-Deep BiLSTM model attained 0.97, 0.98, 0.96, 0.94, and 0.96 for accuracy, F-score, precision, recall and critical success index, respectively when K-Fold is 10. The performance outcome emphasizes the model’s efficacy and accurate prediction and classification of heart disease.