The Application of Genetic Algorithm for Unsupervised Classification of ECG

Abstract

AbstractIn this chapter,we have proposed an integrated methodology for electrocardiogram (ECG) based differentiation of arrhythmia and normal sinus rhythm using genetic algorithm optimized k-means clustering. Open source databases consisting of the MIT BIH arrhythmia and MIT BIH normal sinus rhythm data are used. The methodology consists of QRS-complex detection using the Pan-Tompkins algorithm, principal component analysis (PCA), and subsequent pattern classification using the k-means classifier, error back propagation neural network (EBPNN) classifier, and genetic algorithm optimized k-means clustering. The m-fold cross-validation scheme is used in choosing the training and testing sets for classification. The k-means classifier provides an average accuracy of 91.21 % over all folds, whereas EBPNN provides a greater average accuracy of 95.79 %. In the proposed method, the k-means classifier is optimized using the genetic algorithm (GA), and the accuracy of this classifier is 95.79 %, which is equal to that of EBPNN. In conclusion, the classification accuracy of simple unsupervised classifiers can be increased to near that of supervised classifiers by optimization using GA. The application of GA to other unsupervised algorithms to yield higher accuracy as a future direction is also observed. KeywordsElectrocardiogramPrincipal component analysisNeural networkGenetic algorithmMIT-BIH database