Comparison between Machine Learning Models and Neural Networks on Music Genre Classification

Abstract

In terms of music genre classification, neural networks and machine learning models have their respective advantages. This paper aims to compare the performance and feature extraction capability between neural networks and traditional machine learning algorithms on music genre classification. All the components of 9 main music features, each with seven statistical values, were extracted as essential features, and different dimension reduction methods were applied. This paper compares the performance of training the features by neural networks and machine learning models. Finally, this paper used the output of layers in the neural networks as features and applied traditional machine learning models for training to see if their performance could be optimized. The result showed that the performance was raised by about 20%, compared to the essential features, and raised by about 5%, compared to the reduced features. So, it can be concluded that the feature extraction capability of neural networks is better than traditional machine learning models. Also, using features filtered by neural networks and applying traditional machine learning models for training is a method providing both excellent performance and high efficiency.