ArcGIS Pro Workflow and Python Code for Furtner et al., 2026 - J. Archaeol Method Theory "Evaluating Random Forest Model Performance for Cave and Sinkhole Prediction in the Cradle of Humankind, South Africa: Preliminary Analysis and Variable Importance Assessments"

These files contain the ArcGIS Pro Workflow and Python Code used in the article "Evaluating Random Forest Model Performance for Cave and Sinkhole Prediction in the Cradle of Humankind, South Africa: Preliminary Analysis and Variable Importance Assessments" published in the Journal of Archaeological Method and Theory by Margaret J. Furtner, Robert L. Anemone, Lei Wang, and Juliet K. Brophy.

The ArcGIS Pro Workflow (filename: Furtner_etal_2026_JArchaeolMethodTheory_ArcGISProWorkflow.pdf) contains detailed instructions on the exact tools and tool parameters used to create 48 rasters depicting landscape characteristics of the Cradle of Humankind, South Africa. Raster pixel values overlapping training data points were extracted in table format for input into a Random Forest model.

The Python Code (filename: Furtner_etal_2026_JArchaeolMethodTheory_PythonCode.py) was written for Random Forest model training and evaluation using a spatial 10-fold cross validation method, as well as for performing permutation feature importance on individual variables. Six functions were written for this script, which helped prepare the data for cross validation by splitting the input data into 10 folds based on their site clusters, obtained average evaluation metrics for the 10 training iterations, created an average confusion matrix using the 10 trained models, and performed a permutation feature importance method that separates multicollinear variables.