Support vector machine classifier with feature extraction stage as an efficient tool for atrial fibrillation detection improvement

Intensively explored support vector machine (SVM) due to its several unique advantages was successfully verified as a time series predicting tool in last years. In presented work the improvement of SVM classifier by introduction to a system a preliminary feature extraction stage is proposed. Based on ECG signals from patients suffering from atrial fibrillation (AF) a new feature vector based on separate time, frequency and mixed-domain time-frequency (TF) parameters was created. As a efficient tool for non-stationary signals analysis the discrete wavelet transform was used to obtain the TF signal representation and then new parameters based on energy and entropy measure were computed. Proposed system (FESVM) was tested on the set of 20 AF and 20 patients from control group (CG) divided into learning and verifying subsets. Obtained results showed, that the ability of generalization for enriched FESVM based system increased, due to selectively choosing only the most representative features for analyzed AF detection problem.