A Comprehensive Study in Benchmarking Feature Selection and Classification Approaches for Traditional Malay Music Genre Classification

Machine learning techniques for automated musical genre classification are currently widely studied. With large collections of digital musical files, one approach to classification is to classify by musical genres such as pop, rock and classical in Western music. Beat, pitch and temporal related features are extracted from audio signals and various machine learning algorithms are applied for classification. Features that resulted in better classification accuracies for Traditional Malay Music (TMM), in comparison to western music, in a previous study were beat related features. However, only the J48 classifier was used and in this study we perform a more comprehensive investigation on improving the classification of TMM. In addition, feature selection was performed for dimensionality reduction. Classification accuracies using classifiers of varying paradigms on a dataset comprising ten TMM genres were obtained. Results identify potentially useful classifiers and show the impact of adding a feature selection phase for TMM genre classification.