A Comprehensive Approach to Song Popularity Forecasting

• Published in: 2024 2nd International Conference on Artificial Intelligence and Machine Learning Applications Theme: Healthcare and Internet of Things (AIMLA) pp. 1 - 5
• Main Authors: N, Kishore, P, Santhiya, Asib M, Mohamed Askar, S, Kirubaaharan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15-03-2024
• Subjects: Classification algorithms; Industries; Prediction algorithms; Predictive models; Robustness; Technological innovation; Turning
• DOI: 10.1109/AIMLA59606.2024.10531405

Assessing the potential success of a song based on its acoustic characteristics is a crucial task within the music industry. Traditionally, this has been approached by leveraging audio features independently. In this study, we take a novel approach by jointly exploiting audio features, framing as a regression on prediction task. Our methodology employs a wide architecture, enabling a comprehensive analysis. Additionally, we enhance feature sets by incorporating information about track release years. Evaluation on the Million Song Dataset, using Billboard Hot 100 inclusion as a metric for song success, reveals that our approach surpasses baseline methods and those utilizing low- or high-level features in isolation. Notably, the inclusion of release year alongside mood and vocal features further enhances predictive accuracy, highlighting the efficacy of our comprehensive model in gauging song popularity.