Effective prediction of heart disease using hybrid ensemble deep learning and tunicate swarm algorithm

• Published in: Journal of biomolecular structure & dynamics Vol. 40; no. 23; pp. 13334 - 13345
• Main Authors: Wankhede, Jaishri, Sambandam, Palaniappan, Kumar, Magesh
• Format: Journal Article
• Language: English
• Published: England Taylor & Francis 01.01.2022
• Subjects: Algorithms; CVD; Data Mining; Deep Learning; ensemble deep learning; heart disease; Heart Diseases - diagnosis; Humans; India; UCI dataset; India; CVD; ensemble deep learning; Data mining; UCI dataset; heart disease
• ISSN: 0739-1102; 1538-0254; 1538-0254
• DOI: 10.1080/07391102.2021.1987328

Heart disease (HD) is the major reason for the rampant cause of death around the world. It is deemed as a crucial illness among the middle and old age people which tends to high mortality rates. Recently, Effects of HD is presenting a shocking rise in India. Prediction of HD is considered as the major concern as people are engaged with their day-to-day life and not bothering about their health issues due to the tight schedule of work. Various symptoms may occur for the people who got affected with HD and the recognition of the disease tends to be difficult. Based on the clinical dataset, Data mining techniques are employed for gathering the hidden information. In the present effort, a Hybrid TSA-EDL (Hybrid Tunicate Swarm Algorithm and Ensemble Deep Learning) is implemented for the exact determination of HD. The main tasks indulged for the HD prediction are Pre-processing, clustering and classification. The relevant, irrelevant and redundant features are grouped by DBSCAN (Density-based clustering with noise). At last, the classification process is performed by the hybrid classifier. The proposed work is implemented using the python platform. Two datasets have been included for the analysis as University of California Irvine (UCI) and Cardiovascular Disease (CVD). The different performance metrics used for the analysis are accuracy, recall, specificity, precision, probability of misclassification error, root mean square error, F-score, false positive rate and false negative rate. The obtained performances are differentiated with the outcomes of UCI Cleveland HD dataset and other previous algorithms. As a matter of fact, the performance of the proposed work is increased by attaining the accuracy (98.33%) in CVD and (97.5%) in UCI. Communicated by Ramaswamy H. Sarma