Accurate prediction of underground depth is essential for underground resistive structures
In earthquake-prone areas like Tokyo, where soil liquefaction is high, knowing the depth can help engineers design safer buildings and prevent soil-related disasters.
This purpose uses conventional methods such as standard penetration tests (SPTs). These methods have proven reliable, but are time-consuming, labor-intensive and costly.
Researchers at the Shibaura Institute of Technology (SIT) have demonstrated that machine learning (ML) can provide powerful and cost-effective alternatives. Their research shows that ML not only improves the accuracy of depth prediction, but also allows for scalable and efficient disaster risk assessments in urban areas.
Applying ML to geotechnology
The researchers used a large dataset consisting of 942 geological surveys and SPT records in the Tokyo metropolitan area. They applied three common ML algorithms, Random Forest (RF), Artificial Neural Network (ANN), and Support Vector Machine (SVM), to predict the depth of the bearing layer.
Two scenarios were tested to assess the effectiveness of these models. The first (case-1) used only geographical data such as latitude, longitude, and rise. Second (Case-2) added detailed stratigraphic classification data including the type and structure of the subterranean soil layer.
Of the three ML algorithms, the random forest model consistently provided the most accurate and reliable results. In Case 2, which includes more detailed data, the RF model achieved an average absolute error of just 0.86 meters, significantly outperforming other methods, and even outperforming its own performance in a simpler case-1 scenario (an error of 1.26 meters).
This improvement demonstrates the importance of including stratigraphic data in the ML model of geotechnical applications. Not only does it improve prediction accuracy, it also improves robustness to data noise, which is a key factor when working with real data sets.
The effect of data density on accuracy
Upon further their research, the team investigated how the density of spatial data points affects prediction accuracy. They created six datasets with varying densities ranging from 0.5 to 3.0 data points per square kilometre. The results showed that the prediction accuracy of the RF model improves with higher spatial data density. This indicates that the reliability of ML predictions can be significantly improved when dense data sets are available.
ML-based forecasting models provide practical solutions for urban planners and civil engineers working in seismic areas. Unlike traditional SPT research, ML models require fewer resources and are ideal for large applications.
As computing technology evolves and more geological data is accessible, ML can be integrated into real-time systems for dynamic infrastructure planning. This approach offers promising opportunities for more innovative and safer urban development, especially in earthquake-prone cities such as Tokyo. From building and bridge location optimization to planning underground transportation networks, ML provides versatile tools for resilient infrastructure.
This study illustrates important advances in geotechnical engineering. By combining machine learning with existing geological data, stakeholders can reduce costs, increase safety and streamline planning processes.
