| dc.description.abstract | In the music industry there is a need to analyse the significant features that distinguish highly rated songs from lower rated ones. Then an artist can test their music tracks to check whether it will gain potential popularity, before mass production and if the rating is lower they can focus on the significant features present in popular tracks. Our study address this by developing a machine learning approach to classify music tracks based on user ratings. There were many research performed in the area of music genre classification, music recommendation using vanilla neural networks, recurrent neural networks and convolutional neural networks. The research mainly focuses on the classification of Sinhala songs. Our dataset is consisting of 11,000 Sinhala music tracks each having several attributes. From each track we extract 3 meaningful features. For the feature extraction process we used a python library. The output has three distinct classes that specify the user rating. A Multi-layer neural network was implemented. 500 training epochs with 60 neurons in each hidden layer were used. Initially, with 3031 training tracks and 1299 testing tracks we achieved an accuracy of 86%. With this, we conclude that the development of a multilayer neural network to automate the process of determining the rating for a song is in a successful stage compared with the existing approaches. | en_US |
