Performance analysis of features obtained by PCA (principal component analysis) dimensionality reduction method for diagnosing PAF (paroxysmal atrial fibrillation) patients

Principal Component Analysis (PCA) is an offered scheme for feature extraction and dimension reduction. It has been used extensively in many applications involving high-dimensional data. In this study, we compared the effectivity of PCA features extracted from 33 short-term Heart Rate Variability (HRV) features obtained from normal sinus rhythm (NSR) ECG records for the diagnosis of Paroxysmal Atrial Fibrillation (PAF) disease. Within this framework, different data sets consisting of 33 to 1 features obtained from PCA were used as input to the classification algorithm, which is chosen as the K-Nearest Neighbor (kNN) algorithm. Different values for K and difference distance metrics were utilized to find the best performance. Then the same procedure is applied to another HRV dataset. This set consists of 8 best HRV indices chosen from among the 33 HRV indices by a Genetic Algorithm. The obtained results from both studies elicit that it is possible to further reduce the number of input dimension of a classification system by using PCA algorithm without a reduction in the performance of the system.