An Evaluation of Dimensionality Reduction and Classification Techniques for Cardiac Disease Diagnosis from ECG Signals with Various Deep Learning Classifiers

Abstract

Classification of cardiac diseases from electrocardiogram signals is essential for enhancing patient results to minimize healthcare costs, early detection and accurate diagnosis. This research investigates the integration of dimensionality reduction methods with various deep learning classifiers to improve the accuracy and efficient classification of cardiac disease. Uniform Manifold Approximation and Projection combined with Principal Component Analysis is used for dimensionality reduction, that captures both global and local data structures. Deep learning classifiers with convolutional neural networks, capsule networks, recurrent neural networks, graph neural networks, deep long short-term memory and automatical attention-based convolutional neural networks are employed for classification. The Adaptive spiral Flying Sparrow Search algorithm optimizes classifier parameters for enhance accuracy. Performance is evaluated through various metrics, with area under the receiver operating characteristic curve, accuracy, F1-Score, precision and recall. The proposed method's outcomes are compared with and without optimization to demonstrate the efficiency and attains 92.16%, 96.15%, 91.95%, 94.65%, 91.45% and 92.85% accuracy respectively for each classification method.