Abstract

The design of an intelligent system for identification of cardiac arrhythmias is considered using ECG data set. The arrhythmia identification system that includes feature selection and classification stages is proposed. Cardiac arrhythmias are characterized by many input data. In the paper, a sequential feature selection method is used to reduce the size of the input feature space. The idea behind this study is to find out a reduced feature space so that a classifier built using this tiny dataset can perform better than a classifier built using the original dataset. After feature selection, fuzzy neural networks (FNNs) is applied for the classification purpose. The structure of FNNs is proposed and training algorithm is developed. The classification performance has been evaluated using 10-fold cross-validation. The proposed algorithms have been implemented and evaluated on the UCI ECG dataset. The proposed FNN based approach has provided attractive classification accuracy.

Keywords: Arrhythmia; Multivariate classification; Feature selection; Neural networks

Abbreviations: ECG: Electrocardiogram; SA: Sinoatrial; VFI5: Voting Feature Intervals; FE: Feature Elimination; ROC: Receiver Operating Characteristics; AUC: Area under the Curve; LVQ: Learning Vector Quantization; MSE: Mean Squared Error; SVM: Support Vector Machine; FNN: Fuzzy Neural Network; SFS: Sequential Feature Selection; FCM: Fuzzy C-Means Clustering; SFS: Sequential Forward Selection; SBS: Sequential Backward Selection