Hybrid Transformer-CNN-LSTM Architecture for Improved Cardiovascular Risk Stratification
A Madhumathi, S Mahalakshmi, P Vigneshwaran, Balaji Mohan, S. P. Santhoshkumar, E P Thiruvengadam
Abstract
Exploiting the potentials inherent in deep learning for the prediction of cardiovascular ailments, this research introduces an innovative hybrid framework that amalgamates sophisticated neural network architectures with evolutionary optimization methodologies. By integrating the Tab Transformer alongside Hybrid CNN+LSTM models, the framework aspires to exploit both structured and sequential datasets to forecast heart disease with unmatched precision. By using feature extraction to handle tabular, image, and signal data, Tab Transformer makes it possible to predict heart disease with accuracy. The Cleveland Heart Disease dataset functions as the foundation for the development of predictive models, encompassing essential preprocessing procedures such as the imputation of absent values, encoding of categorical variables, and normalization of continuous features. The fine-tuning of hyper parameters, facilitated by Hyperband and Hyperband optimization, guarantees the identification of optimal learning rates, batch sizes, and model specifications. The findings are scrupulously assessed utilizing a comprehensive set of metrics, with particular emphasis on the Receiver Operating Characteristic (ROC) curve, accuracy, precision, recall, F1-score, and confusion matrices, underscoring the effectiveness of this pioneering approach. In particular, the proposed Tab-Transformer enhanced by Hyperband optimization attained an accuracy of 98.54%, outerperforming XGBoost's 97.57% benchmark a recall rate of 97.09%, a precision of 100%, and an F1-score of 98.52%. This research signifies a substantial advancement in the development of dependable, data-centric solutions for the detection of cardiovascular illness.