Analysis and Discussion of Feature Extraction Technology for Musical Genre Classification

Music Information Retrieval (MIR) applications using deep learning methods are diverse and well-developed. This study utilized the GTZAN dataset, widely used for music genre classification, which includes 10 genres, each with 100 tracks, totaling 1,000 30-second clips, providing diversity and representativeness. The primary purpose of this study is to examine changes in the accuracy of selected music genre classifier programs under different analysis models by reducing feature dimensionality. According to the experimental results, the overall accuracy is higher than the 93.24% used by the original Kaggle platform model provider using KNN. Among them, the number of features of 95 and the number of batches of 128 under the MLP model will produce an accuracy of 95.50%.