An empirical study on mood classification in music through computational approaches

Music Information Retrieval aims computational approaches that can be used to facilitate managing, e.g., indexing, retrieving, storing, etc., music data in large-volume music datasets. In this study, we empirically explore different classification schemes on the mood classification problems in music data. It is long recognized that using mood to classify music data is subjective and ambiguous. Through comprehensive empirical experiments, we demonstrate that the current classification schemes are not sufficient to conduct music classification through mood. Various issues, such as feature selection, feature discretization, etc., are analyzed and discussed. The main purposes of this study is to find through empirical experiments what combinations of classifiers and feature selection techniques work better to classify moods in music data and in the meanwhile, analyze and discuss various issues related to the mood classification problem.