CNN-BiLSTM based model for heart disease classification using deep learning techniques

Abstract

Heart disease remains a main cause of death worldwide, showing the necessity for accurate and timely automated diagnostic systems. Hence, precise and early detection is necessary and is done by some existing machine learning (ML) and deep learning (DL) algorithms. Existing ML and DL models, like convolutional neural network (CNN), long short-term memory (LSTM), CNN-LSTM, and some hybrid approaches, shows good performance. However, they struggle to capture sequential dependencies and spatial feature interactions simultaneously. Further more, hybrid CNN–BiLSTM models to tabular heart disease datasets remains unexplored. To address this gap, this paper, proposes a convolutional neural network–bidirectional long short-term memory (CNN-BiLSTM) hybrid model. This proposed model employs bidirectional processing to improve the temporal features extraction. The proposed hybrid model accurately stores sequential data of the past and the future to do predictions. The experimental validation is done with two different heart disease datasets. The implementation includes learning rate scheduling and hyperparameter adjustment for optimisation. Experimental results on two benchmark datasets demonstrate the proposed model effectiveness. The proposed model got 98.54% accuracy with heart disease dataset and 81.94% accuracy with heart failure clinical dataset, done using 5-fold cross-validation.