Effective cardiac disease classification using FS-XGB and GWO approach

• Cardiovascular disease (CVDs) are the major cause of mortality globally. The electrocardiogram (ECG) is a simple and non-invasive method to monitor the functioning of the heart. Each CVDs are characterized by certain distortion in the ECG. • The utilization of empirical mode decomposition (EMD) to decompose signals into intrinsic mode functions (IMFs), allowing the tracking of even minute changes in the ECG signal. Statistical features extracted from these IMFs are employed for classification. • Reducing the number of features in a machine learning model, referred to as feature selection, is crucial for enhancing model performance, reducing overfitting, speeding up training, and improving interpretability . Implementation of the grey wolf optimization technique to select the optimal subset of features necessary for disease classification, thereby reducing the feature count. • Sorting of features based on their importance and subsequent assignment of regularization term reg_aplha and reg_lambda in the XGB algorithm. This process not only reduces computation time but also further diminishes the number of features, constructing a robust predictive model and enhancing the classification accuracy of the system. • The proposed methodology achieves a maximum classification accuracy , precision, recall, F1-score and AUC value of 98.8 %, 100 %, 99.8 %, 100 % and 98.8 % with just seven optimal features. The number of features used is significantly fewer than the number of features used in existing works. Globally, cardiovascular diseases (CVDs) are a leading cause of death; however, their impact can be greatly mitigated by early detection and treatment. Machine learning (ML)-based algorithms that use features extracted from electrocardiogram (ECG) signals are known to provide good accuracy in predicting various CVDs. Thus, in order to build more effective and efficient machine learning models, it is necessary to extract significant features from ECGs. In order to reduce overfitting and training overhead and improve model performance even more, feature selection or dimensionality reduction is essential. In this regard, the current work uses the grey wolf optimization (GWO) technique to pick a reduced feature set after extracting pertinent characteristics from ECG signals in order to identify five different types of CVDs. On the basis of the feature relevance of the chosen features, a feature-specific extreme gradient boosting approach (FS-XGB) is also suggested. The suggested FS-XGB classifier's performance is contrasted with that of other machine learning techniques, including gradient boosting method, AdaBoost, naïve Bayes, and support vector machine (SVM). The proposed methodology achieves a maximum classification accuracy, precision, recall, F1-score, and AUC value of 98.8 %, 100 %, 99.8 %, 100 %, and 98.8 %, respectively, with just seven optimal features, significantly fewer than the number of features used in existing works.