Article Preview
TopExisting approaches for the automatic detection of cardiac arrhythmia employ different machine learning methods including Support Vector Machines (Asl, Setarehdan, & Mohebbi, 2008), Decision Trees (Exarchos et al., 2007), Artificial Neural Networks (Acharya, Bhat, Iyengar, Rao, & Dua, 2003; Kara & Okandan, 2007), fuzzy classifiers (M. Tsipouras, Goletsis, & Fotiadis, 2004) and ensemble machine learning models (M. G. Tsipouras & Fotiadis, 2004) to name a few.
Acharya et al. (2003), investigated the use of Artificial Neural Networks (ANN) trained via back propagation and a fuzzy classifier to automatically classify heart rate signals into 8 different classes (i.e., normal, pre-ventricular contraction, complete heart block, sick sinus syndrome, left bundle branch block, ischaemic/dilated cardiomyopathy, atrial fibrillation, and ventricular fibrillation). For evaluation purposes, they employed the MIT-BIH arrhythmia database and sampled 1,000 cases for each classification category. The results that they obtained showed that the proposed machine learning models achieve a classification accuracy of approximately 80%-85% with the fuzzy classifier slightly outperforming the ANN.
Exarchos et al. (2007) developed a hybrid machine learning model which integrates decision trees with a fuzzy classifier. The hybrid model firstly uses decision trees to automatically extract association rules between the input features. The association rules are subsequently used as input to the fuzzy classifier. The hybrid classification model was evaluated on two closely related tasks, namely ischaemic and arrhythmic beat detection. Experimental results indicated that the hybrid classification model achieves a very high classification accuracy of 92% and 96% for the ischaemic and arrhythmic beat detection, respectively.