Determining the number of soil layers using the surface wave method based on decision tree and random forest

• Published in: IOP conference series. Earth and environmental science 2024-05, Vol.1337 (1), p.12022
• Main Authors: Zhou, Zan, Lok, Thomas M H
• Format: Article
• Language: English
• Subjects: Algorithms; Civil engineering; Data analysis; Decision analysis; Decision trees; Dispersion curve analysis; Earthquake engineering; Earthquakes; Geophysics; Learning algorithms; Machine learning; Model accuracy; S waves; Seismic activity; Seismic engineering; Soil analysis; Soil dispersion; Soil improvement; Soil layers; Surface waves; Synthetic data; Velocity distribution; Wave dispersion; Wave velocity; Wavelengths
• ISSN: 1755-1307; 1755-1315
• DOI: 10.1088/1755-1315/1337/1/012022

The surface wave method is an efficient non-invasive technique to infer shear wave velocity profiles and has wide applications in civil engineering, earthquake engineering, and geophysics. Inversion analysis is a key step in this method. However, inferring the soil layer number is a challenging problem in inversion analysis and needs improvement. Machine-learning algorithms, including decision tree and random forest, are used in this study to infer the number of soil layers based on the dispersion curve of the surface wave. This process includes two steps for obtaining the model. In Step 1, a large set of synthetic dispersion curves is produced, and in Step 2, a model using machine-learning algorithms is trained based on the synthetic data. The analysis results showed that the obtained model could gain more than 75% accuracy in the test data. Comparing the results from the two machine learning models indicated that the random forest method produced a more favorable prediction. Furthermore, investigating the importance of features implied that the longest and shortest wavelengths were more important than other features.