Music recommendation system based on cosine similarity and supervised genre classification

Categorizing musical styles can be useful in solving various practical problems, such as establishing musical relationships between songs, similar songs, and finding communities that share an interest in a particular genre. Our goal in this research is to determine the most effective machine learning technique to accurately predict song genres using the K-Nearest Neighbors (K-NN) and Support Vector Machine (SVM) algorithms. In addition, this article offers a contrastive examination of the K-Nearest Neighbors (K-NN) and Support Vector Machine (SVM) when dimensioning is considered and without using Principal Component Analysis (PCA) for dimension reduction. MFCC is used to collect data from datasets. In addition, each track uses the MFCC feature. The results reveal that the K-Nearest Neighbors and Support Vector Machine offer more precise results without reducing dimensions than PCA results. The accuracy of using the PCA method is 58% and has the potential to decrease. In this music genre classification, K-Nearest Neighbors (K-NN) and Support Vector Machine (SVM) are proven to be more efficient classifiers. K-Nearest Neighbors accuracy is 64,9%, and Support Vector Machine (SVM) accuracy is 77%. Not only that, but we also created a recommender system using cosine similarity to provide recommendations for songs that have relatively the same genre. From one sample of the songs tested, five songs were obtained that had the same genre with an average accuracy of 80%.