Global Journal of Human-Social Science, B: Geography, Environmental Science and Disaster Management, Volume 22 Issue 3

Clustering of Fine-Grained Tropical Soils using Data Science Tools Applied to their Geotechnical Properties Mayssa Alves Da Silva Sousa α , Roberto Quental Coutinho σ & Laura Maria Goretti Da Motta ρ Abstract - The characterization of fine-grained tropical soils for use in pavements has evolved since the 1980s, however, even today these soils are still discarded or underused in infrastructure works because they do not fully meet the requirements established by traditional classification methodologies or even by the CBR. Tropical soils present peculiarities of geotechnical behavior regarding elastic and plastic deformability, as many authors have already observed. This article contributes to this distinction by analyzing the grouping of thirteen fine-grained soils from northeastern Brazil through the application of data science tools to the results of geotechnical tests. More than fifty geotechnical parameters obtained in the laboratory were considered. By means of simple and multiple linear regressions, they were analyzed in a hierarchical cluster, using Ward's linkage method and Euclidean distance. The results showed that the mechanical behavior of soil compaction and the granulometry, especially the quantities of silt and fine sand, were decisive for the initial division of soils into clusters. For refining the clusters, we considered the different types of geotechnical parameters (physical, chemical and mechanical) in order to highlight other important variables in the distinction of the behavior of these materials. Additionally, microscopic images were compared also using data science techniques and the results corroborated those obtained in cluster analysis, indicating that soil mineralogical characteristics are associated with with the results of laboratory geotechnical tests. Keywords: tropical soils. geotechnical properties. geotechnical behavior. cluster analysis. data science. I. I ntroduction n Brazil, the study of the behavior of fine-grained tropical soils, in the perspective of pavement engineering, began several decades ago, especially with the works of Nogami and Villibor, 1991, 1995), which introduced the classification called MCT. These are soils typical of the tropical environment that, according to the concept adopted by the Committee on Tropical Soils of the International Society of Soil Mechanics and Foundation Engineering - ISSMFE in 1985, present "peculiarities of properties and behavior, in relation to non-tropical soils, due to the performance in the same geological and/or pedological processes, typical of humid tropical regions". It is also emphasized that the introduction of repeated load tests for the determination of the resilient modulus and permanent deformation of this type of soil (Medina and Preussler, 1980; Svenson, 1980) consolidated the appropriate mechanical characteristics of fine-grained tropical soils, complementing the MCT methodology. Due to fine granulation, most of these soils are usually discarded or underused in infrastructure works because they do not present geotechnical parameters that fit the traditional selection criteria ( Transportation Research Board - TRB and California Bearing Ratio - CBR ). However, it has been demonstrated by several surveys (Nogami and Villibor, 1991; Guimarães, 2009; Medina and Motta, 2015; Sousa, 2016; Dalla Roza and Motta, 2018; Lima, Motta and Guimarães, 2017; Lima et al., 2020; Guimarães, Motta and Castro, 2019; Guimarães, Silva Filho and Castro, 2021; among several others) that, regardless of granulometry, consistency indexes and CBR, many of the fine-grained tropical soils have excellent mechanical performance in terms of resilience and plastic deformability, justifying their use in road and railway pavements. Currently, many laboratory tests are carried out to expand knowledge about soil behavior for geotechnical purposes, determining their physical, mechanical, chemical and mineralogical characteristics. The joint analysis of the results obtained by these tests can be performed by means of clustering techniques and can provide valuable information to understand the behavior of soils considering several variables at the same time, by a multivariate analysis. In this context, Frank and Todeschini (1994) define Cluster Analysis as a set of multivariate exploratory methods that seek to find clusters, based on some criterion of similarity between objects (or variables), and that the result of clustering depends greatly on the method used, the standardization of variables and the measure of similarity chosen. The main premise is that the groups or "clusters" formed should be as homogeneous as possible and the differences between the various clusters as large as possible. I © 2022 Global Journals Volume XXII Issue III Version I 9 ( ) Global Journal of Human Social Science - Year 2022 B Author α : Civil Engineering Department, State University of Maranhão, Brazil. e-mail: mayssasousa@professor.uema.br Author σ : Civil Engineering Department, Federal University of Pernambuco, Recife, Brazil. e-mail: roberto.coutinho@ufpe.br Author ρ : Civil Engineering Department, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil. e-mail: laura@coc.ufrj.br