Effect of Feature Selection on The Accuracy of Music Genre Classification using SVM Classifier

Abstract

This research aims to analyze the effect of feature selection on the accuracy of music genre classification using support vector machine with radial basis function kernel as a classifier. In this research, the music dataset from Spotify is used, which is one of the best-selling music streaming platforms today. The selected feature is metadata because it is considered to have simpler processing than audio feature extraction. The music contained in the Spotify dataset also has complete metadata so that the metadata feature can be used properly. At the feature selection stage, some features are combined in different combination groups (FC1, FC2, FC3, FC4). The classification results prove each feature combination has an accuracy result that has a significant difference, where the best accuracy is 80% and the lowest is 67%. Where the combination of FC1 and FC2 features produces the same accuracy of 80%, but because FC2 has a smaller number of features, so the FC2 combination is recommended because with fewer features, so logically the computing time is shorter.