Reproduce “Evolutionary Ensemble Learning for Multivariate Time Series Prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hui Song, A. K. Qin, Flora D. Salim
Researchers Hui Song, A. K. Qin, and Flora D. Salim developed an evolutionary ensemble learning framework for multivariate time series prediction that simultaneously optimizes channel selection, feature extraction, and predictive modeling. The framework, which uses multi-objective evolutionary algorithms and selective ensemble learning, consistently outperformed state-of-the-art methods like LSTM and DA-RNN on real-world electricity consumption and air quality datasets.
Key insights
- Selective ensemble methods (SFS+LS, SBS+LS) consistently improve prediction performance compared to simple averaging or basic least squares combination.
- The proposed framework exhibits reduced sensitivity to algorithmic parameters like population size, addressing a common challenge in evolutionary ensemble learning.
- Leveraging and combining diverse Pareto optimal solutions generated from multiple MOEAs with different population sizes leads to enhanced overall prediction performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.