Evaluation structures for machine learning models in geotechnical engineering

There is currently a lot of interest in applying machine learning (ML) techniques to problems in geotechnical (soil and rock) engineering and adjacent fields such as engineering geology. Recent literature emphasizes the need to focus beyond methodological challenges, and the importance of data centricity, transparency, suitability for practice and geotechnical context – together, the so-called "data-centric geotechnics". This review paper offers additional perspective to be contemplated for successful applications of ML in geotechnics: one should explore and discuss (i) the problem to be solved, (ii) the type, quality and quantity of data, and (iii) the methodology/algorithm. The paper further discusses that more strict guidelines and protocols are required for evaluating data and trained ML models if they are to be accepted and successfully integrated into practice. In the transition to data-centric practices, geotechnical engineering, a traditionally data-poor field, has much to learn from fields where decision-making based on data has a long and rich history.