HYBRID FEATURE EXTRACTION AND STACKING BASED ENSEMBLE CLASSIFIER MODEL FOR CARDIOMYOPATHY CLASSIFICATION
Cardiovascular diseases are considered as a serious challenge globally. Cardiomyopathy is one of the chronic and severe diseases, which is one of the leading causes of human death worldwide. The early prediction of these types of diseases can help to provide a better diagnosis. Recently, machine learning-based models have gained an attraction to classify the cardiomyopathy disease. However, several machine learning-based approaches have been presented during the last decade to detect and classify cardiomyopathy using ECG signal processing. But, due to poor performance of peak and QRS analysis, these techniques fail to achieve the desired performance. Hence, this work focuses on improving peak detection performance and cardiomyopathy classification using machine learning. In order to improve the feature extraction, this work presents a combined model that uses HRV features, QRS complex analysis, heart rate features, and residual features, and generated a feature vector with 96 attributes for each ECG signal. Later, this paper presents a stacking-based ensemble classifier because the single classifier based models fail to handle these irregular attributes. The proposed stacked-ensemble classifier uses SVM, KNN and decision tree as a base learner and linear SVM and meta-classifier. The classification model shows that the proposed approach achieves 95.45% accuracy for cardiomyopathy classification and 99.93% accuracy for ECG peak detection.