Novel hybrid model for music genre classification based on support vector machine
Srishti Sharma, Prasenjeet Fulzele, S. Indu · 2018
Automatic Music genre classification using machine learning techniques has gathered momentum in the recent past and continues to govern this area of research with promising results, saving time and human efforts. Machine Learning techniques have successfully dominated the methods used for the automation of this classification. In this paper, a novel method of music genre classification is proposed based on stacking of Support Vector Machine (SVM) with Relevance Vector Machine (RVM) and Decision Trees. The model uses the acoustic properties of the audio files as their features for classification. The three models are trained as Error-Correcting Output Code Classifiers and they individually classify the audio files with certain posterior probabilities. The results from the three classifiers are fused using the sum rule to evaluate the final outcome of the combined model. The performance of this hybrid model is evaluated on the GTZAN music dataset and is compared with individual performances of the models used in the combination. The proposed combined model outperformed the other models with an accuracy of 87% confirming the efficient utilization of the advantages of individual models.