Classification of User Satisfaction with Spotify Playlist Recommendations Using Random Forest and Hyperparameter Tuning

Music playlist recommendations require an assessment of user satisfaction to evaluate their suitability for listeners’ preferences. This study applies Random Forest to classify user satisfaction with Spotify-recommended playlists and examines the effect of hyperparameter tuning on classification performance. A quantitative experimental approach was used with questionnaire data from 205 respondents, comprising 14 predictors and one satisfaction target with five categories. The dataset was divided into 164 training and 41 test observations. The baseline model used 100 trees, and hyperparameter optimization was conducted using GridSearchCV. The selected configuration comprised 200 trees, the gini criterion, unrestricted maximum depth, a minimum of two samples for splitting, and one sample per leaf. Its mean cross-validation accuracy was 77.47%. Both models achieved 85.37% accuracy, 89.16% weighted precision, 85.37% weighted recall, and 85.16% weighted F1-score, with 35 correct and six incorrect test predictions. The frequency of liking recommended songs was the most important feature, contributing 17.24%. Tuning selected the best validation configuration within the tested search space but did not improve test performance. These findings support satisfaction classification as an input for evaluating playlist recommendations. However, performance for the very dissatisfied category could not be assessed because it was absent from the test set.