Enhancing ECG Signal Classification Accuracy Through Gaussian Modeling Method

Abstract

The automatic classification of ECG signals, also known as computer-based classification, has become critically important in the diagnostic landscape of E-health. This study presents an innovative approach that employs Gaussian Modeling to enhance the accuracy of Beat signal approximation prior to training and classification. Previous methodologies largely relied on Artificial Intelligence (AI) and were contingent on the quality of collected field data sets, which often contained numerous artifacts and noise. This study demonstrates that addressing and eliminating these issues before training can significantly improve classification outcomes. The performance of the proposed approach was rigorously evaluated through several classifiers, including Support Vector Machine (SVM), Decision Trees (DT), K-Nearest Neighbours (KNN), Random Forest (RF), Naive Bayes (NB), Quadratic Discriminant Analysis (QDL), and Convolutional Neural Networks (CNN). These classifiers were applied to the MIT-BIH arrhythmia database, revealing a significant enhancement in results compared to conventional methods. Our findings underscore the efficacy of the Gaussian function in modeling ECG signals, improving the accuracy of various classifiers. Remarkable levels of accuracy, sensitivity, and specificity were achieved across classifiers, with some reaching an accuracy rate of 100%. Notably, the CNN classifier exhibited exceptional performance, demonstrating an accuracy rate of 99.65%, sensitivity of 99.64%, and specificity of 99.88%. This study contributes to the ongoing efforts in the E-health domain to improve diagnostic procedures through AI, offering a significant advancement in ECG signal classification.