Abstract:
This paper presents rule extraction and feature selection system to detect abnormality in ECG signals. Genetic Algorithm-Neural Network (GA-NN) Approach is used to distinguish between the presence and absence of cardiac arrhythmia and perform feature selection. Following this process, rule sets are extracted in order to guide the diagnosis of cardiac arrhythmia. The rule sets are extracted based on selected features because rule extraction without feature selection may result in rules to be more complex than human may realize. C4.5, RIPPER, PART and HotSpot methods are used to perform rule extraction. The ECG dataset used in this study is obtained from UCI Arrhythmia Database. In this dataset, all the anomalies are grouped into one abnormal class and the rest is grouped into one normal class. As a comparison, k-Nearest Neighbor (k-NN), Support Vector Machines (SVM), Naive Bayes and Bayesian Networks have been tested on the arrhythmia dataset. For dimensionality reduction purpose, recursive feature extractor (RFE-SVM), correlation based feature selection (CFS), principal component analysis (PCA) and factor analysis (FA) have been applied. According to test results, GA-NN outperforms other techniques.
