Machine learning techniques for cohesive soil classification in construction in Vietnam

Abstract

Accurate soil classification is imperative for determining land suitability for various construction projects in construction and geotechnical engineering. The physical and mechanical properties of soil significantly influence the design of foundations, the assessment of landslide risks, and the overall stability of structures. Recognizing the limitations of traditional soil classification methods, which are often labor-intensive and time-consuming, this research introduces machine learning as a transformative tool for enhancing soil classification processes. Utilizing K-Nearest Neighbors (KNN) and Support Vector Machine (SVM) algorithms, this study analyzes 5,869 soil samples collected from 39 construction projects in Ho Chi Minh City, Vietnam, to evaluate the efficacy of machine learning techniques in classifying construction soils. The study identifies optimal strategies that significantly improve classification accuracy through a methodical investigation that includes varying training set sizes and integrating directly obtained and indirectly derived soil features. The findings underscore the importance of incorporating liquid and plastic limits and their derived indices, with the KNN model demonstrating superior performance in specific scenarios. This research highlights the potential of machine learning to revolutionize traditional soil classification methods. It provides foundational insights for future advancements in geotechnical engineering, aiming to achieve safer, more efficient, and sustainable construction practices.