Turning Time Series into Algebraic Equations: Symbolic Machine Learning for Interpretable Modeling of Chaotic Time Series.
Under Review., 2026
Chaotic time series are difficult to forecast due to nonlinear dynamics and sensitivity to initial conditions. In this work, we propose two symbolic forecasting approaches, Symbolic Neural Forecaster and Symbolic Tree Forecaster, that learn explicit algebraic equations directly from data. These models combine competitive short-term forecasting performance with interpretability, offering insights into the underlying system dynamics.
Overview of the Symbolic Neural Forecaster (SyNF) and Symbolic Tree Forecaster (SyTF) architectures.