Shakirat Aderonke Salihu, Ibrahim Oluwaseyi Lawal, Oluwakemi Christiana Abikoye, Abdullateef Oluwagbemiga Balogun, Hammed Adeleye Mojeed, Fatima EhenezieUsman-Hamza1and Abimbola Ganiyat Akintola

Keywords: Music, Genres, CatBoost, Music Information Retrieval, Music Recommendations

Abstract

Music genre classification is an essential task in the field of music information retrieval, with numerous applications such as music recommendation systems, playlist generation, and music content organization. This paper focuses on the development of a music genre classification system using the CatBoost algorithm.The CatBoost algorithm is a gradient boosting method that has gained popularity due to its ability to handle categorical features effectively. To implement the music genre classification system,a dataset consisting of audio features and corresponding genre labels was utilized. The audio features include both low-level and high-level characteristics, such as spectral features, rhythmic patterns, and tonal properties. The dataset is preprocessed to extract relevant features and normalize them to ensure consistent representation across different music tracks. The CatBoost algorithm is then employed to train a classification model using the preprocessed dataset and classify them into multiple genres,such as rock, pop, jazz, classical, and hip-hop, among others. The model was evaluated using metrics such as accuracy, precision, recall, and F1 score. The model as an accuracy of 70.99%, while the F1-micro score for the testing dataset was 68.22%. Furthermore, the F1-macro score, which takes into account the average F1-score across all classes, was 71.36% for the training dataset and 68.53% for the testing dataset. The model was also deployed as a web-based application to aid music lovers in having fast access to music genre classification and recommendation.