Classification of Bundle Branch Blocks Using MultilayerPerceptron, K-Nearest Neighbor, and Radial-Basis FunctionPredictors

Nowadays, heart diseases have been the most common cause of death inthe whole of the world, and therefore, developing methods forautomatic electrocardiogram prediction is vital for the clinical diagnosisof heart diseases. Neural networks can be considered a more powerfulmethod than existing methods for classifying medical data. Thus, in thisstudy, artificial neural networks were used to classify the BundleBranch Blocks. In this study, K-Nearest Neighbor (KNN), Radial-BasisFunction Networks (RBF), and Multilayer Perceptron (MLP) (with twodifferent hidden layers, two hidden layers (MLP۲) and three hiddenlayers (MLP۳)) predictors are performed. The total number of samplesfrom the UCI cardiac arrhythmia data set is ۴۵۲, with ۲۴۵ normal and۲۰۷ arrhythmia instances. Finally, the predictors are evaluated throughperformance metrics such as mean squared error, accuracy, sensitivity,and specificity. The results of the study depict that the highest value forsensitivity is for the RBF network (۸۲.۶۵%). In addition, its accuracyand Specificity rates are ۹۱.۳۲% and ۹۷%, respectively which areshown as the best predictor. The accuracy values for MLP-two hiddenlayer, MLP-three hidden layer, and KNN predictors are ۵۹.۱۸%,۷۷.۸۹%, and ۹۰.۶۷%, respectively. The results show that by choosingthe RBF predictor, better results can be obtained in QRS detection.Furthermore, the highest sensitivity value in the RBF predictor and itsfaster learning show the advantage of the RBF network over the rest ofthe applied methods. In addition, through applying noises with differentlevels on a training set, predictors’ accuracies of MLP۲, MLP۳, RBF,and KNN are reduced significantly.

Keywords:

Right Bundle Branch Blocks , MultilayerPerceptron , Radial-Basis Function Networks , K-NearestNeighbor algorithm , Prediction , Performance metrics