Bagged tree classification of arrhythmia using wavelets for denoising, compression, and feature extraction

Abstract

Arrhythmia, also known as dysrhythmia, is a condition involving an irregular heartbeat. A problem in the heart may cause problems in other organs, and as time passes, this will lead to more severe problems. Arrhythmia must be detected at an early stage to prevent such a problem occurring in the heart. Detection of arrhythmia from an electrocardiogram is an easy method that does not need much equipment and does not harm the patient. The purpose of this research is to find a faster and more accurate system to classify nine classes of arrhythmia. The St. Petersburg Institute of Cardiological Technics 12-lead arrhythmia database was used for training and testing. Data were compressed and preprocessed (denoising, trend elimination, baseline correction, and normalization) before being sent to the system for feature calculation. The wavelet coefficients that displayed the most significant effect on classification were chosen and used as features. Standard deviation and variance were also added to the feature set. Later, principal component analysis (PCA) was used to reduce the number of features further. After deciding the features, the performance of the basic classification methods and spiking neural network was checked to determine whether there was a better classifier to be used for our research. Tenfold cross-validation was applied to the training dataset. Bagged trees were found to produce better results. The classifiers' performance was tested by sensitivity, specificity, and accuracy.