Hierarchical SVM-kNN to Classify Music Emotion

Abstract

The emotional component in a music classification is more powerful than the others. This research addresses a music emotion classification. A hierarchical classification system using a Support Vector Machine (SVM) and a k-Nearest Neighbors (kNN) is proposed. The experiments using 120 pop-rock music data with the emotional label based on the AllMusicGuide website split into four classes: "Happy", "Angry", "Sad", and "Relax" show that the proposed hierarchical model is capable of increasing the absolute performance of music emotion classification by 19.33% in the SVM (Kernel: RBF) and 13.33% in the kNN (k = 5). The best combination three-level classifier, the arrangement of the three best classifiers for each level in hierarchical music emotion classification is by using the SVM (Kernel: Linear) classifier at Level 1, then kNN (k = 3) at Level 2.1 and Level 2.2.