A Comparative Analysis of Traditional Machine Learning, Deep Neural Network, and Statistical Modeling for Music Genre Classification
Vishnu Vardhan Kollipara, V. Hari Shankar Dinesh, Priyanka D. Kumar, T. Deepika, M. Srinivas · 2025
A range of general features of music, such as Mel-Frequency Cepstral Coefficients and spectrograms, is used to describe musically descriptive confines of different genres. While such a comparison is made between traditional machine learning models, like SVMs and random forests, with the more elaborative work on deep-learning architectures like CNNs, RNNs, and probabilistic techniques like Bayesian networks, which capture uncertainty and perhaps provide a more fine-grained prediction. Deep-learning models, especially CNNs, appear to achieve high accuracies in music genre classification over the GTZAN dataset. Probabilistic models, however, have the advantage of handling noisy or incomplete data by enforcing some meaningful interpretation of the underlying relations between features and genres. This research provides a thorough comparison of these techniques and aids in improving the music genre classification systems.