Indonesian Journal of Electrical Engineering and Computer Science Vol. 12, No. 3, December 2018, pp. 984~994 ISSN: 2502-4752, DOI: 10.11591/ijeecs.v12.i3.pp984-994  984 Journal homepage: http://iaescore.com/journals/index.php/ijeecs Comparison of Multiscale Entropy Techniques for Lung Sound Classification Achmad Rizal 1 , Risanuri Hidayat 2 , Hanung Adi Nugroho 3 1,2,3 Department of Electrical Engineering & Information Technology, Universitas Gadjah Mada, Jl Grafika no 2, Mlati, Sleman, D.I. Yogyakarta, Indonesia 1 School of Electrical Engineering, Telkom University Jl Telekomunikasi no 1, Ters. Buah Batu, Bojong Soang, Bandung, Indonesia Article Info ABSTRACT Article history: Received Apr 27, 2018 Revised Aug 24, 2018 Accepted Oct 8, 2018 Lung sound is a biological signal that can be used to determine the health level of the respiratory tract. Various digital signal processing techniques have been developed for automatic classification of lung sounds. Entropy is one of the parameters used to measure the biomedical signal complexity. Multiscale entropy is introduced to measure the entropy of a signal at a particular scale range. Over time, various multiscale entropy techniques have been proposed to measure the complexity of biological signals and other physical signals. In this paper, some multiscale entropy techniques for lung sound classification are compared. The result of the comparison indicates that the Multiscale Permutation Entropy (MPE) produces the highest accuracy of 97.98% for five lung sound datasets. The result was achieved for the scale 1-10 producing ten features for each lung sound data. This result is better than other seven entropies. Multiscale entropy analysis can improve the accuracy of lung sound classification without requiring any features other than entropy. Keywords: Multiscale entropy Lung sound Coarse-grained procedure Multilayer perceptron Copyright © 2018 Institute of Advanced Engineering and Science. All rights reserved. Corresponding Author: Achmad Rizal, Department of Electrical Engineering & Information Technology, Universitas Gadjah Mada, Jl Grafika no 2, Mlati, Sleman, D.I. Yogyakarta, Indonesia. Email: rizal.s3te14@mail.ugm.ac.id 1. INTRODUCTION Lung sound is one of the biological signals that emerge from the respiration process. Any changes in it are generated from foreign bodies or physiological changes in the respiratory tract caused by diseases [1]. Differences in patterns of lung sounds can be heard by a doctor using a stethoscope to diagnose diseases [2]. Auscultation technique, on the other hand, is very subjective for being dependent upon existing experience and expertise of the doctor. Various techniques for analyzing lung sounds using computers have been developed. Some of these techniques include time-domain analysis techniques, such as statistical analysis based on Hjorth descriptor [3], empirical mode decomposition (EMD) [4], or fractal analysis [5]. Several researchers proposed to perform lung sound analysis in the frequency domain, such as quantile vector frequency [6] or MFCC [7]. Meanwhile, wavelet analysis has been used in [8] for classifying abnormal lung sounds. One of the popular biological signal analysis methods is multiscale entropy (MSE), proposed by Costa et al. with a coarse-grained procedure for multiscale process and sample entropy for entropy measurement [9]. Subsequently, several variants of MSE emerge such as refined-MSE, composite-MSE or adaptive MSE (AME) [10]. Other researchers modified their entropy measurement techniques, which result in multiscale permutation entropy (MPE) [11], multiscale approximate entropy (MApEN) [12], and multiscale fuzzy entropy [13]. In the case of lung sounds, multiscale entropy has been used to analyze the