Music Genre Classification: A Comparative Analysis of CNN and XGBoost Approaches with Mel-frequency cepstral coefficients and Mel Spectrograms

In recent years, various well-designed algorithms have empowered musicplatforms to provide content based on one's preferences. Music genres aredefined through various aspects, including acoustic features and culturalconsiderations. Music genre classification works well with content-basedfiltering, which recommends content based on music similarity to users. Given aconsiderable dataset, one premise is automatic annotation using machinelearning or deep learning methods that can effectively classify audio files.The effectiveness of systems largely depends on feature and model selection, asdifferent architectures and features can facilitate each other and yielddifferent results. In this study, we conduct a comparative study investigatingthe performances of three models: a proposed convolutional neural network(CNN), the VGG16 with fully connected layers (FC), and an eXtreme GradientBoosting (XGBoost) approach on different features: 30-second Mel spectrogramand 3-second Mel-frequency cepstral coefficients (MFCCs). The results show thatthe MFCC XGBoost model outperformed the others. Furthermore, applying datasegmentation in the data preprocessing phase can significantly enhance theperformance of the CNNs.