Prediction of Artists' Rankings by Regression
Felipe Lopes de Melo Faria, Álvaro R. Pereira Jr., Luiz Henrique de Campos Merschmann · Journal of the Association for Information Systems · 2015
The construction of rankings consists of ordering retrieved results according to certain criteria. Rankings can provide relevant information to analysts from different sectors of industry. For the music industry, rankings enable understanding how musical genres and popularity of artists and their songs evolve over time, allowing analyses of history data and trends. Due to the importance of building rankings in the musical scope, data mining techniques have been used to predict rankings by using information from social media. This work evaluates regression models for prediction of artists’ rankings using historical data (daily rankings of artists) extracted from website Vagalume. Three regression techniques (k-Nearest Neighbors - k-NN, Multiple Linear Regression - MLR and Random Forests - RF) were evaluated in this study considering different scenarios. Results obtained from experiments showed that predictions with low error rates can be obtained, indicating that data mining techniques can be used to obtain information to assist the music industry in decision making.