Methodological and Reporting Practice in Geotechnical Machine Learning: An Audit of 149 Studies

Methodological and Reporting Practice in Geotechnical Machine Learning: An Audit of 149 Studies This record contains the supporting materials for a systematic audit of machine-learning (ML) practice in geotechnical engineering. A stratified random sample of 149 studies was drawn from fifteen geotechnical journals (2022–2025) and coded against eleven classes. The classes are grouped into methodological errors that can bias reported performance (Tier 1), evaluation weaknesses (Tier 2) and reporting deficits (Tier 3). One author coded all 149 studies, and a second author independently coded a random subset of 30. Four studies that released their data were re-evaluated under structure-respecting validation protocols. Contents protocol_and_codebook.md: audit protocol and codebook, as originally deposited. Deviations from it are listed in Table 2 of the paper. coding_sheet.xlsx: primary codes for all 149 studies, identified by DOI (tab "Coding"); the second coder's codes for 30 studies (tab "Second coder"); screening log; codebook reference. sample_master.csv: bibliographic metadata (authors, title, year, journal, DOI) for the 150 sampled records. compute_statistics.py: reproduces the prevalence estimates, confidence intervals, trend and subdomain analyses, prevalence by journal, and agreement between coders. make_figures.py: reproduces Figures 1 and 2. geo071_site_characterisation.py, geo100_settlement_surrogate.py, geo103_pffp_classification.py, geo133_classification.py: re-evaluation scripts. table4_regression_folds.py, table4_geo103_baselines.py, table4_geo133_baselines.py: fold-level statistics, confidence intervals and baselines for Table 4. fig1–fig4 (PNG): figures as they appear in the paper. README.txt: file descriptions, data sources, software versions and changes from the previous version. Re-evaluation data The datasets used in the re-evaluations belong to the original studies and are not included here. They are available from the repositories cited in those studies. Exact links, commit identifiers and retrieval dates are listed in README.txt. Software: Python 3.12.3, pandas 3.0.2, NumPy 2.4.4, SciPy 1.17.1, statsmodels 0.15.0, scikit-learn 1.8.0, Matplotlib 3.10.8.