AUTHOR=Liu Huifen , Lin Peiyuan , Wang Jianqiang 

TITLE=Machine learning approaches to estimation of the compressibility of soft soils

JOURNAL=Frontiers in Earth Science

VOLUME=Volume 11 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/earth-science/articles/10.3389/feart.2023.1147825

DOI=10.3389/feart.2023.1147825

ISSN=2296-6463

ABSTRACT=The modulus of compression and coefficient of compressibility of soft soils are key parameters for assessing deformation of geotechnical infrastructure. However, consolidation tests to determine these two indices are very time consuming and the results could be easily and largely influenced by the workmanship, testing apparatus and many other factors. Therefore, developing a simple approach to accurately estimate the compressibility indices is of great interest. This study presents the development of three machine learning (ML) models for mapping the two compressibility indices for soft soils. The ML models are artificial neural network (ANN), random forest, and support vector machine. A database containing 743 sets of measured physical and compression parameters of soft soils is adopted to train and validate the models. The accuracies of the ML models are statistically evaluated using bias factor defined as the ratio of measured to predicted compression indices. Results showed that the three ML models are all accurate on average with low dispersion in prediction accuracy. The ANN model is the best as it gives simple analytical form and has no hidden dependency between bias and predicted indicies. Last, the probability distribution functions of the bias factors are also determined using the fit-to-tail technique.