In the Paper Predicting the Origin of Music [1], it is described how Random Forest regression (RFR) and k-Nearest Neighbours are used to predict where non-western music originates. This paper achieved significant results, yet the explanation of the algorithms remain unclear. It was thought that perhaps some insight into human culture could be extracted from the algorithms. This paper investigates which music classification features are influential in determining the geographic origin of world music. By using the machine learning algorithms K-Nearest Neighbour Regression, Random Forest Regression and XGBoost, the results of the original paper have been reproduced using the dataset provided [1] and the feature importance has been inspected. For the REDI course at Univerity of Twente, Sharon Engbers and I found that the most influential music features are features which are outside the sound spectrum of tones which are most important to humans, but seem to have more to do with the music recording equipment. Also some more questions and topics for further research regarding machine learning on sound and geological data are proposed.
[1] F. Zhou, C. Q and R. D. Kink, “Predicting the Geographical Origin of Music,” in 2014 IEEE International Conference on Data Mining, Shenzhen, 2014.