Machine Learning in Geotechnical Practice: From Data to Regulatory Compliance

  This workshop addresses the practical challenges of applying machine learning to geotechnical engineering problems. Participants will examine critical technical challenges, including transforming subjective geological and geotechnical assessments into robust ML training data, reconciling ML predictions with Eurocode requirements when training data derives from empirical correlations with limited validation, and quantifying and communicating model uncertainty to practitioners, communities, and regulators.

The workshop will explore data preprocessing techniques and validation strategies when ground truth is ambiguous or unavailable. Special focus will be given to distinguishing between aleatory uncertainty (which ML can quantify through statistical methods) and epistemic uncertainty arising from fundamental knowledge gaps that statistical processes cannot capture. We will examine when ML genuinely helps reduce uncertainty versus when it merely provides computational complexity without improving predictive reliability.

Participants will gain practical frameworks for implementing ML in geotechnical practice while maintaining transparency about model limitations. The session covers strategies for deploying ML-based approaches within regulatory frameworks, effectively communicating uncertainty to diverse stakeholders, and identifying scenarios where traditional empirical methods remain more appropriate than ML approaches. Case studies will demonstrate both successful applications and cautionary examples of ML applications.