ESTIMATION OF CLAY SWELLING PROPERTIES IN KARAGANDA TERRITORY USING MACHINE LEARNING METHODS

Abstract

The relevance of the study is caused by the need to identify the dependence of the clays swelling on their physical characteristics in order to reduce time and money resources during geotechnical surveys. The active development of the construction industry leads to the development of territories composed of soils, which, as a result of moisture, increase in volume – swell. The main aim of the study is to establish the relationship between the relative swelling of clay rocks and their physical characteristics, the determination of which requires minimal resources using machine learning methods. Objects: Quaternary and Neogene clays of Karaganda, which compose the geological section of the territories used for the construction of buildings and structures. Methods: creation of laboratory database (physical and compression characteristics, particle size distribution) and field (description of soils: color, presence of inclusions, determination of groundwater level, sampling intervals) studies of clays in excel; application of the high-level programming language «Python» to develop mathematical models using the «Anaconda» distribution kit; the Pareto theorem application for training and validation of the resulting model; use of the «Mean Squared Error» indicator to assess the adequacy of the developed models. Results. Three predictive models of the relative clay swelling were developed. The laboratory and geological parameters of 103 clay samples taken as a result of geotechnical surveys in Karaganda, Kazakhstan, were the input data. The following machine learning algorithms were used: Random Forest, Multilinear regression, Support vector machines. According to the «Mean Squared Error» criterion, the Random Forest model was chosen to develop a relative swelling model.

Keywords

• clay rocks
• clays
• swelling
• random forest model
• quaternary deposits
• neogene deposits