Reliability and Regional Reproducibility of Hyperparameter-Optimized Landslide Susceptibility Mapping in Two Contrasting Loess Regions-data

This dataset provides the processed data, evaluation outputs, and source code supporting a comparative study of hyperparameter-optimized landslide susceptibility mapping in Xifeng District and Longxi County, Gansu Province, China. The study examined conditioning factor selection, predictive performance, susceptibility map evaluation, spatial sensitivity, and hyperparameter optimization efficiency across two contrasting loess regions. The modeling datasets contain 214 samples in Xifeng, comprising 107 landslide and 107 non-landslide samples, and 294 samples in Longxi, comprising 147 samples of each class. Fourteen candidate conditioning factors were considered in Xifeng and 15 in Longxi. Sample identifiers, class labels, training and test assignments, and cross-validation fold assignments are provided. Six models—logistic regression, support vector machine, random forest, multilayer perceptron, gated recurrent unit, and one-dimensional residual network—were evaluated with budgeted grid search, random search, the tree-structured Parzen estimator, and a genetic algorithm under a common 50-trial budget. The repository includes multicollinearity diagnostics, candidate factor sets, Shapley additive explanations, optimization summaries for all 96 regional model–factor-set–method combinations, selected model configurations, and available trial logs. Model evaluation files contain held-out predictions, receiver operating characteristic curve coordinates, statistical comparisons, repeated-training results, and out-of-fold predictions for random and spatial cross-validation. Map evaluation files include 12 continuous susceptibility rasters, 12 classified rasters, success rate and prediction rate curve coordinates, class-area statistics, the seed cell area index, and enrichment ratios. Source code, software environment specifications, a data dictionary, and SHA-256 checksums accompany the data. These materials support verification of the reported statistics and reuse of the processed modeling samples. Sample coordinates are withheld for spatial confidentiality, while susceptibility rasters retain their geographic reference. Original inventory coordinates and full-grid conditioning-factor inputs are not included. Complete individual-trial trajectories are unavailable for the Longxi optimization runs; task summaries and available logs are supplied, with their scope documented in the README files.