Comparison of Neural Networks and XGBoost Algorithm for Music Genre Classification

The purpose of this research paper is to implement Neural Networks and XGBoost based models in order to generate music genre classifiers & analyze their performance. This analysis is based on their accuracy, speed and various other factors. The dataset is taken from the kaggle website. The dataset includes audio files, mel spectrogram & CSV with extracted features. For the Neural Network (NN) model, audio files are used as the dataset. This dataset is analyzed thoroughly and different waveforms are plotted using the python package called Librosa. The dataset is preprocessed and its MFCC features are extracted. The MFCC values are treated as the input and their respective music genres are supposed to be the output for the Neural Network model. The model is trained with four neural network layers. Similarly for the XGBoost algorithm, CSV with extracted features is taken as the main dataset. Some fraction of the dataset is trained with the model and for the rest of the data is analyzed for comparison between the two. The accuracy/ error plot of test/training data and a confusion matrix, is generated for both and models. Models are trained using open-source libraries-Tensorflow, Keras, Xgboost, Sklearn, etc