Estimating soil SPT N-values from investigative drilling data: correlation and machine learning analyses based on field drilling tests

Abstract Investigative drilling (ID) is a variant of measurement while drilling (MWD) techniques that has been successfully applied to multiple site investigation projects in Australia; however, no reliable correlation or model currently exists for estimating Standard Penetration Test (SPT) N-values from drilling data due to the noise and uncertainty in industrial MWD data. This study investigates the feasibility of estimating soil SPT N-values from ID data through field drilling tests and statistical and data-driven analyses. A purpose-designed comparative drilling program was conducted in a uniform, clay-rich ground to minimise spatial variability and acquire high-quality paired datasets from ID and conventional drilling. Both conventional MWD indices and machine learning were employed to correlate the drilling data with SPT N-values. Among the three selected MWD indices, the Somerton index showed the strongest correlation with SPT N-values ( r = 0.57), although its applicability was limited to shallow depths due to bogging effects in clay. In contrast, the Random Forest model significantly improved predictive performance ( R 2 up to 0.894), demonstrating the capability of machine learning to capture complex, non-linear relationships between drilling parameters and soil resistance. Data partitioning was found to significantly influence model performance due to the small dataset, whereas the training process of Random Forest has a limited effect. Feature importance analysis identified depth as the dominant predictor, while penetration rate became the most influential parameter when depth was excluded. The study highlights both the potential and limitations of ID-based SPT N-value prediction, providing a foundation for future development of reliable, data-driven site investigation methods.