Performance Comparison of Various Classifier Algorithms on Bioacoustic Sound Data Using Principal Component Analysis

Abstract

Recognition of the sounds recorded in natural environments is more difficult than recognition of the sounds recorded in the laboratory or artificially generated. Especially the sounds taken from similar individuals is a factor that make the recognition difficult. In this study, machine learning techniques were compared in order to recognize different family, genus and species of anuran sounds recorded in natural environments. The data set named Anuran Calls (MFCCs) from the UCI database was used. This data set includes Mel Frequency Cepstral Coefficients of anuran sounds taken from natural environments. On this the data set, the performances of Logistic Regression, Support Vector Machines and Decision tree classifiers were compared. In addition, with Principal Component Analysis, dimension reduction was made in the data set and the changes in classifier performances were compared.