Analisa Audio Features dengan Membandingkan Metode Multiple Regression dan Polynomial Regression untuk Memprediksi Popularitas Lagu

Abstract

Songs are artistic works that expresses ideas and emotion in the forms of rhythms, melodies, and harmonies. Songs are the source of huge profit for musicians or artists from commercial view-point. Based on the data from IFPI, the earnings from the music industry in 2019 reached US$20.2 billion, in which 56.1% of them came from streaming revenue. Spotify is one of the largest and most well-known streaming services in the world today.

This research aims to make predictions of popularity from each song according to the audio feature data taken from Spotify's API. The process of prediction will use 2 regression methods, which are Linear Regression and Polynomial Regression. The model will be made using those 2 methods and will be tested with the R2, Adjusted R2, MAE, and MSE metric systems.

From the analysis of the implementation to the program, the Linear Regression method had garnered the average results as follows: 0.23614 for R2, 0.23536 for Adjusted R2, and had average errors 17.38129 for MAE method, 442.31700 for MSE method. Using the Polynomial Regression method, the average results were: 0.31496 for R2, 0.25880 for Adjusted R2, and had average errors 16.47367 for MAE method, 409.76242 for MSE method.