A pruned fuzzy k-nearest neighbor classifier with application to electrocardiogram based cardiac arrhytmia recognition

This paper renders a fuzzy nearest neighbor classifier with data pruning to reduce the number of stored prototypes to minimize memory and computational time requirements. The incorporation of fuzzy set theory into nearest neighbor classification makes the decision process more flexible and adaptable to noise in the data. We have also embodied an efficient approach for nearest neighbor search in our algorithm which results in significant reduction in computational time during training and classification. We present results of classification of different data sets from the University of California, Irvine (UCI) machine learning repository to illustrate the effectiveness of the suggested approach for classification purposes. We also give an application of the proposed classification methodology to electrocardiogram (ECG) based recognition of 9 types of arrhythmias using wavelet domain features. The results obtained (~97% accuracy), clearly indicate the effectiveness of this algorithm in the design of a practical ECG analyzer.