Predicting Music Popularity Using Machine Learning Algorithm and Music Metrics Available in Spotify

Abstract

The exponential growth of online music streaming has given birth to many new platforms among which, the widely used platform is Spotify. The most popular music streaming app's data can be used to predict the capability of a song to be popular before its release with the help of attributes like loudness, energy, acousticness, etc. which is defined when the song is being made. This study helps to predict the popularity of the song using the song metrics available in Spotify by applying Random Forest classifier, K-Nearest neighbour classifier and Linear Support Vector classifier to compare which of these models can effectively predict the popularity. The results found that Random Forest works the best for predicting popularity with high accuracy, precision, recall and F1-score.

Effect of Feature Selection on the Accuracy of Music Popularity Classification Using Machine Learning Algorithms

“…Which of these algorithms predicted popularity effectively was determined by looking at the accuracy, precision, recall, and F1-score metrics. When the results were examined, it was observed that the random forest algorithm gave the best result in estimating popularity [23].…”

Unsupervised Heterogeneous Graph Neural Network for Hit Song Prediction through One Class Learning

“…The musical representation used to define a hit song has different features. We noted works such as [Pareek et al 2022] that use previously processed high-level audio features, [Zangerle et al 2019] that utilized low-level audio features as a base to learn a deep representation, or some approaches that use pre-processing resources as Principal Component Analysis [Ge et al 2020] or autoencoders [Martín-Gutiérrez et al 2020]. The lyrics concentrate on important musical information and are also explored in [Singhi and Brown 2015;Martín-Gutiérrez et al 2020].…”

“…OCL is relevant for the hit song prediction task because, considering the real music market scenario, we have few hit songs in relation to non-hit. This problem has been highlighted in MIR [Ge et al 2020;Martín-Gutiérrez et al 2020;Bertoni et al 2021;Pareek et al 2022] for hit-song prediction. However, it is explored as a problem with a dependency of non-hit songs.…”

“…Generally, the hit song prediction problem is solved by binary learning, where the objective is to predict whether a song will be a hit or non-hit by learning from past hit and non-hit songs [Herremans et al 2014;Pareek et al 2022]. However, the scope of non-hit songs is wide since exists more nonhit songs than hits (imbalanced scenario), so it's hard to cover that scope and label non-hit songs.…”