VISUAL ANALYSIS AND MACHINE LEARNING–BASED DETECTION OF CARDIAC ABNORMALITIES FROM ECG SIGNALS

CVDs has been identified as one of the leading causes of mortality in the world and it is important to make sure that cardiac abnormalities are treated at an early age to enhance the survival of patient. Electrocardiography represents a non-invasive test that is used to measure the activity of the heart and determine abnormal heart rhythm. Hand reading of the ECG signals is however time consuming and highly skilled in skills especially where a significant amount of physiological information is required to be read. As of relatively recent advances in machine learning and biomedical signal processing, nowadays, automated diagnostic systems capable of detecting cardiac abnormalities with pretty high degree of accuracy can be developed. This article study suggests a single-balanced framework of detecting abnormal patterns of heart beats by visual inspection and machine learning techniques on ECG signals. The proposed solution includes signal preprocessing methods in order to remove noise and a line-drift, followed by features extraction methods that reflect the time-domain characteristics and frequency-domain characteristics of ECG signals. The visual analytics software will be utilized to represent the pattern of waveforms and identify the characteristics of the dataset to understand the variations of the signals better. They used several classification models including Support Vector Machines, Rand Forest classifiers, Convolutional Neural Networks, as well as, Long Short-Term Memory networks to evaluate the applicability of machine learning models in detection of abnormalities in ECG signals. According to the results of the experiments, the performance of the deep learning models is superior to the classical classification algorithms and achieves the accuracy of more than ninety-seven percent. The visualization methods are also helpful in the knowledge regarding the model behavior through the support of demonstrating the performance of classifications and characteristics of data. The framework developed depicts the effectiveness of visual-level analysis and machine learning of automated ECG signal interpretation. Such systems can assist clinicians in detecting cardiac abnormalities with less effort and allow developing intelligent healthcare monitoring systems that can potentially carry out cardiac diagnostics in real time.