Electrocardiogram (ECG) heart disease diagnosis using PNN, SVM and Softmax regression classifiers

In this paper, an automatic method is proposed to classify heart beats into 15 classes mapped to five main categories. Dynamic segmentation strategy is utilized to keep into consideration the heart rate variation. Discrete Wavelet Transform (DWT) is then applied on the segmented heart beats to extract the features for the description of each segment. The features extracted are then subjected to principle component analysis (PCA) to remove the irrelevant features due to its high dimension. Thereafter, Support Vector Machine (SVM), Softmax regression and Probabilistic Neural Network (PNN) algorithms are applied to the reduced features. Finally, MIT-BIH dataset is utilized as an evaluation dataset with tenfold cross validation strategy to achieve 97.1%, 98.3% and 93.3% overall accuracy and 88.7%, 83.3% and 42.0% average accuracy using SVM, PNN and Softmax regression respectively.