Chemical compositions of grinded obsidian artefacts from Early Eneolithic site in Nitra-Selenec (Slovakia) and reference samples of carpathian obsidians

Data analysis of chemical composition (ED-XRF) of obsidian artefacts from Nitra. 

The chemical analysis of obsidian artifacts from Nitra was conducted using the R programming environment, applying a variety of statistical techniques and machine learning models. The preprocessing steps included importing raw data from an Excel sheet using the openxlsx package. Red obsidian samples were excluded, and elemental concentrations were normalized using pairwise log-ratio (pwlr) transformation with the compositions package. Missing values were imputed using the zCompositions package, employing the multinomial KM (mKM) approach suitable for compositional data.

Subsequent transformations ensured consistency in elemental sums across samples. Samples with SiO₂ content below 50% were filtered out using the dplyr package. The final dataset underwent standardization of major oxides and trace elements to achieve consistent scaling before classification.

For provenance classification, a Random Forest model from the randomForest package was applied. The model underwent cross-validation to optimize performance, and variable importance was assessed through built-in feature ranking functions. The results were visualized using scatter plots and ternary diagrams generated with the ggplot2 and ggtern packages.

This work was supported by the Slovak Research and Development Agency under the Contract no. APVV-20-0521, APVV-23-0282 and project VEGA 2/0033/23. The research was conducted under the wider project Ready for the future: understanding long-term resilience of the human culture (CZ.02.01.01/00/22_008/0004593) and Institutional grant of Faculty of Science, Masaryk University, no. 2222/315010.