Robustness of Support Vector Machine-based Classification of Heart Rate Signals

Abstract

In this study, we discuss the use of support vector machine (SVM) learning to classify heart rate signals. Each signal is represented by an attribute vector containing a set of statistical measures for the respective signal. At first, the SVM classifier is trained by data (attribute vectors) with known ground truth. Then, the classifier learnt parameters can be used for the categorization of new signals not belonging to the training set. We have experimented with both real and artificial signals and the SVM classifier performs very well even with signals exhibiting very low signal to noise ratio which is not the case for other standard methods proposed by the literature.

Supervised Learning - A Systematic Literature Review

“…In [17], SVM were used to classify and compare the breathing patterns of patients undergoing weaning trials. The authors of [18] employed this algorithm to classify the heart rate signal. They obtained the reading from three distinct sets of individuals: the young, teenagers, and elders.…”

Electrocardiogram Pattern Recognition and Analysis Based on Artificial Neural Networks and Support Vector Machines: A Review

“…Osowski et al [61] HOS, Hermite expansion Arrhythmia detection Kampouraki et al [62] HRV time domain features HRV classification Mohebbi et al [63] RR time and frequency features AF detection Melgani et al [64] Morphology + timing Arrhythmia detection Mehta et al [65] Entropy QRS detection Kostka et al [66] Wavelet AF detection Mehta et al [72] ECG slope P T detection separability among classes. This implies the maximization of a cost-function based on eigenvectors of S w -1 S b .…”

“…For HRV analysis, Kampouraki et al [62] used the Fantasia Database from Physionet (see Section 10) for classifying young and elderly people's HRV while watching a Disney film, through conventional HRV features (section 6). Kampouraki et al [62] used SVM for RR series classification and found that SVM performed better than ANN. They used conventional statistical HRV features, autocorrelation, Shannon entropy, autoregressive coefficients and discrete wavelet transform.…”

“…The vast existing literature on automatic ECG classification can be grouped, for the specific aim of this review, to different categories depending on the applied features and on the classification methods. The large variety of features used to represent ECG can be grouped in terms of features used for (a) arrhythmia detection [6,14,[57][58][59][60][61][62][63][64][65][66][67], (b) rhythm analysis [15,[22][23][24][25][26][27][68][69][70], and (c) human identification . Classification methods range from ANN [12,[59][60][33][34], SVM [61-62, 64-65, 67-68, 70-72], linear discriminant analysis (LDA) [73][74], k-nearest-neighbour (k-NN) [75], mixture of experts (MOE) [76], Bayesian networks [77], Kalman Filtering (KF) [78][79][80][81][82][83][84][85][86][87], Hidden Markov Models (HMM) [88]…”

“…The main challenges for an ECG-based computer-aided-diagnosis system are ECG pre-processing (noise and artifacts removal), QRS detection, heartbeat classification, rhythm analysis, and disease diagnosis. This review does not consider pre-processing and QRS detection as they are very well established areas; instead, it is focussed on pattern recognition methodologies applied to heartbeat classification which is still an area of intense research [61][62][63][64][65][66][67][68][69][70][71][72][73][74][75][76][94][95][96][97]. This review does not consider pattern recognition approaches to ECG noise removal, QRS detection, and ECG segmentation (detection of the main wave-boundaries).…”