Advanced Ensemble Learning Framework for Beats-Per-Minute Prediction Using Stacked Gradient Boosting Models

View at publisher Summarise with VIRAT
Authors: JiaYi Meng
Cited by: 0
References: 5

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Accurate prediction of beats-per-minute in music tracks is crucial for music information retrieval, playlist generation, and audio analysis applications. This paper presents an innovative ensemble learning framework that combines multiple gradient boosting algorithms through a sophisticated stacking methodology with ridge regression as the meta-learner. The proposed approach integrates LightGBM, XGBoost, CatBoost, and Histogram-based Gradient Boosting regressors with comprehensive feature engineering strategies including polynomial interactions, log transformations, and categorical binning. Through rigorous 15-fold cross-validation on a dataset comprising 524,164 training samples with 9 acoustic features, the ensemble model achieves a root mean squared error of 26.46 on validation data, demonstrating superior performance compared to individual base learners. The stacking architecture employs out-of-fold predictions as meta-features, effectively capturing complementary information from diverse gradient boosting implementations while maintaining model generalization through regularization. An ablation study confirms that specific engineered features including logarithmic transformations of TrackDurationMs and polynomial interactions between Energy and RhythmScore provide the most significant performance improvements, while also identifying redundant transformations that can be eliminated without compromising accuracy.

Fields of Study & Subjects

Music and Audio Processing Music Technology and Sound Studies Machine Learning and ELM

Publication details

Journal / source
International Conference on Electrical Automation and Artificial Intelligence (ICEAAI)
Pages
785–791
Publication date
1 January 2026
Language
EN