Classification of Soil Based on Laboratory Tests by UsingClassification Algorithms in Machine Learning

New models based on machine learning classifiers were presented in this paper that can be applied to various problems regarding geotechnics and civil engineering. Kernel SVM and K Nearest Neighbors are types of non-linear artificial intelligence (AI-ML) algorithms which are powerful methods for regression and classification. This study used these two kinds of algorithms to classify soil types. There is no doubt that defining the type and characteristics of soil samples is necessary in tunnelling and civil engineering projects for evaluation the mechanical properties of soil. Considering this topic's importance, ۱۴۳ samples of measured data of the actual project (data from Line ۷ metro, E-W section, Tehran) were used to design the new networks based on machine learning algorithms. The dataset includes plasticity index, gravel content, liquid limit, fines content, and sand content to evaluate and define the type of soil classification. During the pre-construction and construction of tunneling the samples were collected and tested in laboratory with high accuracy and the actual types of soil was obtained. Two types of machine learning classifiers were developed according to the actual condition of soil along the tunnel. The outcomes of classifications were considered for different samples of soil including coarse materials and fines materials. The developed networks showed that they can be considered as powerful models to classify the different types of soil. There were just eight incorrectly identified samples when testing the kernel SVM model (۴۲ samples) and ten incorrectly identified samples when testing the K Nearest Neighbors model (۴۲ samples). Therefore, by using these networks, geotechnical projects can be more precise and have fewer financial and cost problems.

کلیدواژه ها:

MACHINE LEARNING ، GEOTECHNICS ، SOFT GROUND TUNNELLING ، EPB-TBM.