Systematic Benchmarking of ANN, CNN, and LSTM Models for Heart Disease Prediction Using Multimodal Clinical Features

Accuracy and early prediction of heart disease are essential for efficient clinical decision-making and risk management. This paper presents a deep learning-based predictive approach for heart disease diagnosis using a real-time clinical dataset Ayush Multi Speciality Hospital and Research Centre (AMSHRC) Pvt. Ltd. Vijayapura, Karnataka, India augmented with sophisticated diagnostic attributes. The proposed approach combines systematic data preprocessing, feature extraction, and three different neural network architectures: Artificial Neural Network (ANN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) networks, to compare the prediction accuracy and robustness. Five-fold cross-validation stratified and independent test-set validation are performed. Cross-validation results show that CNN yields the highest validation accuracy (96.19% +0.61), MCC (0.9087 +0.0144) and ANN yields equal validation accuracy (95.87% +0.82%), highest recall value (94.92% +1.36%) and highest AUC (0.9915 +0.0020). The performance on test-sets also verifies that ANN is the most appropriate model overall and has an accuracy of 96.20 and AUC of 0.9936. The inclusion of ECG, ECHO, TMT, and CAG features is a very good approach and helps in improving the prediction accuracy, and it also verifies that ANN is a good model that can be used in practice. Key words: Cardiovascular disease (cvds), long short-term memory (lstm), deep learning (dl), convolutional neural network (cnn), artificial neural network (ann), heart disease prediction