Predicting Hit Songs Using Audio and Visual Features
C.-H. Lee, Yi‐Ning Tu · 2025
Factors contributing to a song’s popularity are explored in this study. Recent studies have mainly focused on using acoustic features to identify popular songs. However, we combined audio and visual data to make predictions on 1000 YouTube songs. In total, 1000 songs were grouped into two categories based on YouTube view counts: popular and non-popular. The visual features were extracted using OpenCV. These features were applied using machine learning algorithms, including random forest, support vector machines, decision trees, K-nearest neural networks, and logistic regression. Random forest performed the best, with an accuracy of 82%. Average accuracy increased by 9% in all models when using audio and visual features together. This indicates that visual elements are beneficial for identifying hit songs.