Harmony in Algorithms: Exploring Music Genre Classification Through Machine Learning

Music genres serve as classifications that categorize music according to shared traditions and practices. These classifications can heighten the pleasure derived from music by offering listeners a framework to organize and comprehend diverse musical styles. When employed effectively, genre categorization aids in deepening our comprehension of this art form, identifying innovation, and ultimately refining our ability to assess quality. The primary objective of this endeavor is to explore the distinct characteristics exhibited by various musical genres through their spectral representations, aiming to develop an automated classification system. By curating a meticulously classified music dataset, such as the GTZAN Music Genre dataset, the extracted feature-map of the data is inputted into a neural network model for evaluation. This evaluation includes assessing the accuracy across training, testing, and validation stages, coupled with endeavors to minimize validation losses to a significant extent. Additionally, this paper conducts a comparison of multiple classification algorithms, revealing that CNN achieved an accuracy of 79%, outperforming other methods like SVM, KNN, Naive Bayes, and Random Forest.