On the Classification of ECG Signals Subject to Various Degrees of Dimensionality Reduction

• Published in: E-Health and Bioengineering Conference (Online) pp. 1 - 4
• Main Authors: Fira, Monica, Goras, Liviu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 29.10.2020
• Subjects: Classification algorithms; classifications; Compressed sensing; Data visualization; dimensionality reduction; Electrocardiography; Laplace equations; Laplacian Eigenmaps; Locality Preserving Projections; Principal component analysis; Support vector machines
• ISSN: 2575-5145
• DOI: 10.1109/EHB50910.2020.9280248

In this paper we investigate the classification performances of ECG signals subject to various degrees of dimensionality reduction. For each case, results obtained with several classification algorithms are presented and discussed. Two of the three methods for dimensionality reduction investigated in this work are Laplacian Eigenmaps (LE), Locality Preserving Projections (LPP). The third one, somehow different, is Compressed Sensing (CS) a method for acquiring and reconstructing signals from a reduced number of random projections under the hypothesis of sparsity. The purpose of the analysis is to investigate the advantages and disadvantages of each method and to determine to what degree it can be considered being optimal for a case of dimensionality reduction. The evaluation of the dimensionality reduction effect for the ECG signals was made based on the classification of the signals in the reduced spaces. Moreover, the classification rates of the original signals were compared with the classification rates of the small-dimensional signals using several classifiers.