Classification of ECG signals using LDA with factor analysis method as feature reduction technique

M. Kaur,A. Arora
DOI: https://doi.org/10.3109/03091902.2012.702851
2012-10-17
Journal of Medical Engineering & Technology
Abstract:The analysis of ECG signal, especially the QRS complex as the most characteristic wave in ECG, is a widely accepted approach to study and to classify cardiac dysfunctions. In this paper, first wavelet coefficients calculated for QRS complex are taken as features. Next, factor analysis procedures without rotation and with orthogonal rotation (varimax, equimax and quartimax) are used for feature reduction. The procedure uses the ‘Principal Component Method’ to estimate component loadings. Further, classification has been done with a LDA classifier. The MIT-BIH arrhythmia database is used and five types of beats (normal, PVC, paced, LBBB and RBBB) are considered for analysis. Accuracy, sensitivity and positive predictivity are performance parameters used for comparing performance of feature reduction techniques. Results demonstrate that the equimax rotation method yields maximum average accuracy of 99.056% for unknown data sets among other used methods.
What problem does this paper attempt to address?