From Rhythm to Recognition: “Leveraging Deep Learning and Machine Learning Models for Music Genre Classification”
Authors
Kodilkar, Kartik
Issue Date
2025.17.12
Degree
Master of Business Administration
Publisher
Dublin Business School
Rights holder
Rights
Open Access
Abstract
Music powerfully reflects and shapes human culture, yet the task of classification of music into distinct genres is inherently challenging given the existence of stylistic overlaps, cultural specificity, and the subjectivity of listener interpretation. This research explores the evolving landscape of music genre classification through application of machine learning and deep learning techniques, with a specific focus on the implementation and performance evaluation of Convolutional Neural Networks (CNN), Support Vector Machine (SVM) and Random Forest (RF) algorithms. An extensive evaluation of 30 CNN architectures were carried out, along with 5 SVM and 5 RF model configurations using large datasets like FMA-large and MuMu datasets. Feature extraction techniques – particularly Mel-frequency Cepstral Coefficients (MFCCs) and Mel-spectrogram were determined to be highly significant in distinguishing the acoustic feature specific to each genre. The research addresses persisting challenges to genre classification, such as class imbalance issues, the emergence of hybrid genres, and questions of cultural representation. Through empirical testing and architectural improvements, the research determines strong model configurations that outperforms conventional approaches on genre recall and macro-average F1 score metrics. This research contributes to the development of genre classification systems that are not only technologically adept but also contextually through, offering important implications for applications in music streaming, recommendation systems, and global music archiving.