Music Genre Classification Using Modified K-Nearest Neighbor (MK-NN)

Klasifikasi Genre Musik Menggunakan Modified K-Nearest Neighbor (MK-NN)

Abstract

The genre of music is a grouping of music according to their resemblance to one another and commonly used to organize digital music. To classify music into certain genres, one can do it by listening to the music one by one manually, which will take a long time so that automatic genre assignment is needed which can be done by a number of methods, one of which is the Modified K-Nearest Neighbor. Modified K-Nearest Neighbor method is a further development of its former method called KNearest Neighbor method which adds several additional processes such as validity calculations and weight calculations to provide more information in the selection class for the testing data. Research to find the best H value shows that the H = 70% of the training data is able to produce an accuracy of 54.100% with K = 5 and the proportion ratio of test data and training data is 20:80 (fold 5). The best H value is then used for further testing, which is to compare the K-Nearest Neighbor method with the Modified K-Nearest Neighbor method using two different proportions of test data and training data and each proportion of data also tests a different K value. The results of the classification comparison of the two methods show that the Modified K-Nearest Neighbor method, with the highest accuracy of 55.300% is superior to the K-Nearest Neighbor method with the highest accuracy of 53.300%. The two highest accuracies produced in each method were obtained using K = 5 and the proportion ratio of test data and training data is 10:90 (fold 10).