Study of feature selection algorithms to improve arrhythmia detection performance on ECG signal

Arrhythmia, a potentially life-threatening heart rhythm disturbance, requires accurate diagnosis and treatment. Detection through electrocardiogram (ECG) signals is the most prevalent method, yet the complex analysis often generates irrelevant features, impacting fault detection. This study innovatively addresses these challenges by introducing an effective feature selection algorithm aimed at identifying the most informative features within ECG signals to enhance arrhythmia detection performance. The approach incorporates a reengineering method, involving the analysis and redesign of existing systems to boost their functionality. Utilizing the MIT-BIH ECG signal dataset, a comprehensive compilation from various medical centers, the research encompasses signal processing, feature extraction, and the application of Information Gain (IG). Evaluation metrics unveil the algorithm’s accuracy, with IG implementation achieving an impressive accuracy rate of 98.79%. Moreover, the algorithm demonstrates a specificity of 99.41% and a sensitivity of 98.79%. This feature selection algorithm not only attains high accuracy levels but also presents a novel contribution in aiding doctors to make more precise diagnoses and select appropriate treatments for patients with arrhythmias. The study stands out for its holistic approach, combining feature selection with reengineering techniques, resulting in an innovative methodology for improved arrhythmia detection. Furthermore, the study introduces a novel integration of Information Gain with a Convolutional Neural Network (CNN) architecture, addressing overfitting issues and enhancing the robustness of arrhythmia detection in ECG signals. This integration represents a unique contribution to the field, providing an innovative solution for accurate arrhythmia detection.