A Classification Based Approach to the Prediction of Song Popularity

Music is a continuously evolving field that has existed for centuries as a form of relaxation and entertainment. The field of music and its industry have grown considerably over the last few decades and constant efforts are being made to make hit songs so as to maximize the revenues generated. Online music streaming platforms have come into existence in the last couple of years and have become the most popular methods of streaming music. These platforms provide ways to evaluate the popularity of songs through rankings which are calculated by the number of streams and other factors. While different studies have been carried out to predict the success of songs, this research focuses on using the metadata of the songs and performing profanity and sentiment analysis on the lyrics to predict its popularity. Six machine learning algorithms (Random Forest Classifier, SVM, Decision Tree Classifier, K-Nearest Neighbors, Logistic Regression and Naïve Bayes) were compared and the one with the best accuracy was used.