Optimization of an Automatic Music Genre Classification System via Hyper-Entities

This paper aims at presenting a âcomputational costâ optimization method in an Automatic Music Genre Classification system. In such systems, the training and validation database is often enormous. Consequently, a system based on a nearest neighbor classifier suffers from high computational cost during the classification process. In such cases, a training instance clustering (per class) can be used. Instances (entities) that belong to the same class and have similar feature values are themselves clustered so as to form a hyper-entity. Thus, any new entity pending classification is compared only to the database hyper-entities and the hyper-entities are responsible for deciding in which class the new entity belongs.