Work

FEG-Pro: Supplementary Software for Forecast-Error Growth, Lyapunov-Related Profiling, and Nonlinear Time-Series Feature Analysis

Year: 2026
Type: software
Abstract: This reproducibility archive accompanies the revised manuscript FEG-Pro: A Forecast-Error Growth Framework for Finite-Horizon Instability and Feature Extraction in Nonlinear Time Series. It provides the FEG-Pro k-nearest-neighbor forecast-error-growth implementation and Windows launchers, original/refined IAAFT surrogate and noise-dependent classification experiments, autocorrelation and forecast-error-distribution-entropy analyses, classical Rosenstein/Kantz/TISEAN comparison controls, fitting- and sampling-sensitivity checks, and the regenerated Mackey–Glass benchmark. The Lorenz-63 component includes 13 archived 100,001-point scalar trajectories generated using deterministic RK4 integration and a 900-configuration grid search that selected one common horizon setting against known reference exponents. The archive distinguishes descriptive finite-horizon forecast-error slopes from validated asymptotic Lyapunov exponents. The reference-guided Lorenz calibration is an in-sample sensitivity analysis, not an independently validated accuracy gain. Source data, selected numerical results, run configurations, checksums and provenance notes are included for audit and reproduction.
Language: English
Open Access status: green
OpenAlex ID: W7225896224
Cites:
Cited by:
Related to: 10