MATLAB Code for Substorm Forecasting Using a Machine Learning Approach

This repository contains the MATLAB code and derived result tables supporting the manuscript “Substorm Forecasting Using a Machine Learning Approach.” The MATLAB workflow processes 1-min solar-wind and geomagnetic-index measurements, prepares published substorm-onset lists, constructs balanced substorm and quiet-time datasets, trains Long Short-Term Memory (LSTM) classification models, compares prediction performance across multiple substorm-onset catalogs, performs leave-one-feature-out analysis, evaluates sensitivity to forecast lead time, and generates the publication figures. The input parameters used in the analysis are solar-wind velocity Vx, interplanetary magnetic-field components Bx, By, and Bz, proton density Np, SYM-H, and SML. The study interval is 1998–2007. The record includes the MATLAB analysis and figure-generation scripts together with the individual CSV result tables used to generate the figures and report the numerical results in the manuscript. These result tables include the onset-list comparison, lead-time analysis, leave-one-feature-out results for each onset catalog, and summary feature-importance tables. The underlying solar-wind, geomagnetic-index, and substorm-onset source data are not redistributed in this record. The included documentation provides the required data sources, expected directory structure, preprocessing steps, and script execution order needed to reproduce the analysis from the original data.