Machine learning techniques for cohesive soil classification in construction in Vietnam - KNN and SVM study

Aug 3, 2026

Accurate cohesive soil classification underpins foundation design, geotechnical risk assessment, and structural stability in construction projects. Traditional classification standards rely on laboratory particle-size distribution and Atterberg limits, making workflows labor-intensive and time-consuming. This research introduces machine learning to streamline soil classification for Vietnamese construction sites. Using 5,869 soil samples from 39 projects in Ho Chi Minh City, it compares KNN and SVM while varying training sizes and soil feature sets. The results show that liquid and plastic limits and derived indices are especially influential, with KNN achieving stronger performance in targeted scenarios.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages20

Reading Time50 min

AccessFree to download

Introduction

Machine learning for soil classification

Soil classification standards and limitations

Machine learning methods and related work

Machine learning techniques for cohesive soil classification in construction in Vietnam Danh Thanh Tran1*, Dinh Xuan Tran1, Vinh Hoang Truong1 1Ho Chi Minh City Open University, Ho Chi Minh City, Vietnam *[Corresponding author: danh.tt@ou.edu.vn](Corresponding author: danh.tt@ou.edu.vn) | ARTICLE INFO | ABSTRACT | | --- | --- | | DOI:10 .46223/HCMCOUJS. | Accurate soil classification is imperative for determining | | tech.en.15.2.3816.2025 | 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. | | Received: October 25th, 2024 | Utilizing K-Nearest Neighbors (KNN) and Support Vector | | Revised: December 15th, 2024
Accepted: December 17th, 2024 | 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 | | Keywords: | machine learning to revolutionize traditional soil classification | | geotechnical engineering; KNN; machine learning; soil | methods. It provides foundational insights for future advancements in geotechnical engineering, aiming to achieve | | classification; SVM | safer, more efficient, and sustainable construction practices. | 1. Introduction Soil classification is fundamental in the construction industry, providing critical insights for foundation design, risk assessment, and project cost estimation. In construction, soils are primarily categorized into cohesive and non-cohesive types. In Vietnam, soil classification commonly follows the TCVN 9362:2012 standard (Cong thong tin dien tu Bo Xay dung, 2012), alongside other internationally recognized systems such as USCS (Unified Soil Classification System) , AASHTO (American Association of State Highway and Transport Officials), and ASTM (American Society for Testing and Materials) . These classification systems generally rely on particle size distribution and Atterberg limits to determine soil properties (Casagrande, 1948; Das & Sobhan, 2013) . Obtaining these values necessitates laboratory experiments, including sieve and sedimentation tests for particle composition, tests for moisture content, and determining liquid and plastic limits. Despite their critical importance, these conventional methods are often time-consuming and laborintensive, leading to a demand for more efficient approaches. The rise of Artificial Intelligence (AI) across various sectors, from autonomous vehicles and facial recognition systems to virtual assistants and content recommendation systems, has opened up new possibilities for geotechnical engineering. Early AI applications in this field addressed diverse challenges, such as soil parameter prediction (Mollahasani et al., 2011; Nguyen et al., 2020; Pham et al., 2020; Pham, Mahdis, et al., 2021; Zhang, Wu, et al., 2021), pile load-bearing capacity estimation (Momeni et al., 2020; Pham et al., 2022; Singh & Walia, 2017; Tran et al., 2024), retaining wall design (Ghaleini et al., 2018; Gordanet al., 2019; Koopialipoor, Murlidhar, et al., 2020), TBM op

FAQs

Why is cohesive soil classification critical in construction projects?

Cohesive soil classification provides essential inputs for foundation design, landslide risk evaluation, and overall structural stability, directly affecting safety and project decisions.

What traditional data do conventional soil classification methods use in this context?

They rely on laboratory-derived particle size distribution and Atterberg limits, including sieve and sedimentation tests as well as liquid and plastic limit measurements.

Which machine learning approaches are compared, and how are they evaluated?

The study evaluates K-Nearest Neighbors (KNN) and Support Vector Machine (SVM) using 5,869 samples, testing different training set sizes and combining directly obtained and indirectly derived soil features.

Aug 3, 2026

Upload by Valentina