1*Research Scholar, Mangalayatan University, Aligarh, UP, India. Email: saroj.cse10@gmail.com
2Professor, Mangalayatan University, Aligarh, UP, India. Email: meena.chaudhary@mangalayatan.edu.in
3Professor, AIML, AIT-CSE, Chandigarh University, Mohali, Punjab, India. Email: raghav.mehrain@gmail.com
Because CVD is among the leading causes of death worldwide, early risk prediction is particularly important. This work employs a mixed dataset and differentiates between three types of variables, clinical/lifestyle, and ECG-based ones, aiming at analyzing four DLMs namely- MLP, LSTM, CNN, and ViT. The case for the relevance to PTB Using the ECG portion of the PTB-XL and combining it with UCI Heart Disease features; this resulted in 22 multimodal predictors down the first level. We evaluated the model performance using accuracy, precision, recall, F1-score, and AUC. The results indicate that all architectures perform well (Accuracy ≥ 0.97, AUC ≥ 0.99). LSTM obtained the optimal overall balance (Accuracy = 0.99, Recall = 0.98, F1 = 0.98), though CNN had slightly lower recall, (Recall = 0.97), and ViT reached perfect accuracy (1.00) with slightly lower recall (0.90). SVM joint analysis of three modalities as an example, the hybrid method has shown the feasibility of deep learning in early CVD diagnosis and decision support by means of boosting robustness and clinical relevance as compared to single-modality studies.
Keywords: Cardiovascular Disease, Electrocardiogram (ECG), Deep Learning, Hybrid dataset, Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Vision Transformer (ViT).
How to cite this article: Kumari S, Chaudhary M, Mehra R. Deep Learning Models Performance Analysis for Cardiovascular Disease Using an ECG Based Dataset. Int J Drug Deliv Technol. 2026;16(2): 87-97; DOI: 10.25258/ijddt.16.2.12
Source of support: Nil.
Conflict of interest: None