Auto-FS-Cardiac: Optimizing ECG Heartbeat Classification with Automated Feature Selection using TPOT Template Framework

Cardiovascular diseases (CVDs) constitute a significant global health concern with a profound impact on mortality rates. Recent advancements in artificial intelligence (AI) have facilitated the successful application of automated classification methods for cardiac arrhythmias. This paper introduces 'Auto-FS-Cardiac,' an innovative automated feature selection model. Leveraging Automated Machine Learning (Au-toML) and the Tree-based Pipeline Optimization Tool (TPOT) framework, the model constructs a classification pipeline aimed at distinguishing between five distinct heartbeats in electro-cardiogram (ECG) data sourced from the MIT-BIH database. The study evaluates the performance of Auto-FS-Cardiac under both automated and predefined human-expert feature selection scenarios. Additionally, a comparative analysis with traditional feature selection models provides insights into the proficiency of Auto-FS-Cardiac in generating optimal pipelines for precise ECG heartbeat classification. Auto-FS-Cardiac performance, achieved an accuracy level of 0.9569 with a rapid execution time of 1.9857 seconds. Notably, when utilizing predefined features, the model maintains a consistent accuracy score of 0.9522, albeit with a longer execution time of 14.7836 seconds. This highlights the model's adaptability in balancing high accuracy and efficiency when autonomously managing the feature selection process. The observed tradeoff between efficiency and interpretability suggests that interventions in feature selection may impact these factors.