Implementasi Random Forest untuk Memprediksi Popularitas Lagu di Spotify: Studi Kasus Fitur Audio dan Metadata

The fast-growing digital music industry, especially music streaming platforms like Spotify, has created the need to build models that can predict how popular a song will be. This study aims to use the Random Forest algorithm to predict song popularity based on a mix of audio features and metadata. The dataset was collected using crawling techniques from the Spotify API, with 2,416 songs and features like loudness, tempo, valence, artist name, and release date. The data went through several preprocessing steps, including cleaning, encoding, normalization, and feature selection. The model was tested using different data split ratios (70:30, 80:20, 60:40, and 50:50) to see which gave the best results. The 80% training and 20% testing split showed the best performance, with an MAE of 0.1176, RMSE of 0.01526, and R-squared of 0.3279. The analysis of feature combinations also showed that pairs like name_artist_valence, loudness_acousticness, and length_acousticness had a strong effect on song popularity. This study concludes that the combination of music features and artist identity has a bigger impact on popularity than the release date.