Derived data and reproducibility package for remote early detection of tomato late blight by laser speckle sensing

This dataset accompanies the manuscript Remote early detection of tomato late blight by laser speckle sensing and interpretable machine learning The archive contains the derived feature table and reproducibility materials used for the reported plant-level classification analyses. It includes 720 speckle-video records from 40 tomato plants, corresponding to 20 inoculated and 20 mock-inoculated plants imaged at 6, 18, and 24 h after treatment. For each video, the package provides 105 engineered features: 26 raw speckle/pixel descriptors and 79 phase-correlation-derived position/displacement features. The package includes the combined feature table, plant-held-out fold definitions, frozen model hyperparameters, plant- and video-level predictions, derived statistics tables, and a standalone script that reproduces the main reported XGBoost results. Raw speckle-pattern videos are not included because of their size, approximately 500 GB, and are available from the corresponding author on reasonable request. The accompanying software record contains the code for feature extraction, model training, evaluation, and figure generation.