Prediksi Peringkat Mingguan Lagu Pada Spotify Amerika Serikat Menggunakan Multiple Charts Dataset Dengan Berbagai Metode

Abstract

In 2020, the majority of the music industry's revenue, 62.1%, came from streaming music. As a result, many music business parties are striving for a hit song, particularly on Spotify US chart. However, this is difficult to achieve because nowadays, a song's performance is determined by its performance on various music charts, not by its quality. Due to that, a study in the field of hit song science will be conducted to forecast weekly song ranking on Spotify US using data from Spotify, Shazam, Airplay, and TikTok charts. Multipler linear regression, polynomial regression, gradient boosting tree, and random forest are the methods used in this study to create models, and each model will be compared using adjusted r-squared and mean absolute error (MAE) as evaluation metrics. Random forest produced the best model, with adjusted r-squared and MAE values of 93.133% and 11.687, respectively. The usage of music attribute had a negative impact on model performance. Shazam chart, on the other hand, has been shown to have a positive impact on model performance. Meanwhile, neither the Airplay nor the TikTok charts have a definite positive or negative impact. However, both have been shown to have a very weak relation with model performance. Overall, the dataset combination of Spotify, Shazam, Airplay, and TikTok chart produced the best model in this study.