An ensemble learning feature selection and heterogeneous fusion XGBoost algorithm for arrhythmia classification

The intelligent classification of electrocardiogram (ECG) signals is of great significance in the early diagnosis and intervention of cardiovascular diseases. However, challenges such as feature redundancy and data imbalance often result in reduced classification accuracy. To address these issues, this paper proposes a novel Ensemble Learning Feature Selection (ELFS) and Heterogeneous Fusion XGBoost (HF-XGBoost) algorithm for arrhythmia classification. Compared with conventional machine learning approaches, the proposed algorithm offers the following improvements. Firstly, an ELFS method integrating Bootstrap aggregating (Bagging) ensemble learning with multiple feature selection methods is designed to select more relevant and informative feature subsets, which enhances the discriminative information of features while reducing dimensionality. Secondly, heterogeneous base learners are incorporated into the XGBoost algorithm to effectively capture local information, global gradient information, and high-dimensional sparse representations, which enhances the algorithm ability to exploit minority samples information and mitigates the performance bottlenecks inherent in single algorithm. Furthermore, Bayesian optimization is employed to fine-tune the algorithm hyperparameters to achieve optimal performance. This paper follows the recommendations of the Association for the Advancement of Medical Instrumentation standard and adopts the inter-patient evaluation scheme. Experimental is evaluated on four types of ECG data from the MIT-BIH arrhythmia database, achieving overall precision, specificity, and F1 scores of 97.70%, 93.26%, and 97.57%, respectively, with an accuracy of 97.53%. Compared with other state of the art method, it exhibits superior and more balanced performance across all evaluation metrics.