MetaCost XGBoost Training and Evaluation Dataset with MATBLAB Codes and files for generating proxies

The dataset consists of two curated subsets designed for the classification of alteration types using geochemical and proxy variables. The traditional dataset (Trad_Train.csv and Trad_Test.csv) is derived directly from the original complete geochemical dataset (alldata.csv) without any missing values and includes original geochemical features, serving as a baseline for model training and evaluation. In contrast, the simulated dataset (proxies_alldata.csv) was generated through custom MATLAB scripts that transform the original geochemical features into proxy variables based on multiple geostatistical realizations. These proxies, expressed on a Gaussian scale, may include negative values due to normalization. The target variable, Alteration, was originally encoded as integers using the mapping: 1 = AAA, 2 = IAA, 3 = PHY, 4 = PRO, 5 = PTS, and 6 = UAL. The simulated proxy data was split into the simulated train and test files (Simu_Train.csv and Simu_Test.csv) based on encoded details for the training (=1) and testing data (=2). All supporting files—including datasets, intermediate outputs (e.g., PNGs, variograms), proxy outputs, and an executable for confidence analysis routines are included in the repository except the source code, which is on GitHub Repository. Specifically, the FinalMatlabFiles.zip archive contains the raw input files alldata.csvused to generate the proxies_alldata.csv, it also contains Analysis1.csv and Analysis2.csvfor performing confidence analysis. To run the executable files in place of the .m scripts in MATLAB, users must install the MATLAB Runtime 2023b for Windows 64-bit, available at: https://ssd.mathworks.com/supportfiles/downloads/R2023b/Release/10/deployment_files/installer/complete/win64/MATLAB_Runtime_R2023b_Update_10_win64.zip. Details on the input files for confidence analysis: Analysis1.csv and Analysis2.csv These files contain two columns for the test data: column 1 = match or mismatch between predicted and true alterations? column 2 = probability of a correct classification, according to bootstrapped samples (Analysis1.csv) or to simulated proxies (Analysis2.csv)