🤖 AI Summary
Addressing the challenges of hyperparameter-dependent tuning, limited generalizability, and poor interpretability in high-fidelity time-series forecasting of chaotic systems, this paper proposes TreeDOX—a parameter-free tree-based model. TreeDOX innovatively couples time-delayed embedding (to explicitly capture short-term memory) with an ensemble of extremely randomized trees (Extra-Trees) regressors, enabling unsupervised feature dimensionality reduction and end-to-end prediction within a reconstructed phase space. Crucially, it requires no hyperparameter optimization, ensuring robustness, full interpretability, and plug-and-play deployment. Evaluated on canonical chaotic benchmarks—including the Hénon map, Lorenz system, and Kuramoto–Sivashinsky equation—as well as real-world Southern Oscillation Index data, TreeDOX achieves state-of-the-art prediction accuracy, significantly outperforming both mainstream deep learning approaches and classical dynamical modeling techniques.
📝 Abstract
Model-free forecasting of the temporal evolution of chaotic systems is crucial but challenging. Existing solutions require hyperparameter tuning, significantly hindering their wider adoption. In this work, we introduce a tree-based approach not requiring hyperparameter tuning: TreeDOX. It uses time delay overembedding as explicit short-term memory and Extra-Trees Regressors to perform feature reduction and forecasting. We demonstrate the state-of-the-art performance of TreeDOX using the Henon map, Lorenz and Kuramoto-Sivashinsky systems, and the real-world Southern Oscillation Index.