Master's dissertation · Aeronautics Institute of Technology (ITA) · 2018

Application of Machine Learning Techniques for Soil Classification from CPTU

Lucas Orbolato Carvalho · Advisor: Dimas Betioli Ribeiro

Dissertation

Master of Science in Aeronautical Infrastructure Engineering, field of Airport Infrastructure, Instituto Tecnológico de Aeronáutica (ITA), São José dos Campos, Brazil. Final document dated December 27, 2018.

Signed dissertation PDF Publications Author homepage

Overview

This dissertation develops a multidimensional machine-learning framework for soil classification from cone penetration test (CPT/CPTU) data. The work combines supervised and unsupervised learning, evaluates raw and normalized CPT measurements, examines the contribution of depth and geological age, and applies data cleaning, balancing, transformation, statistical comparison, and 10-fold cross-validation.

The supervised study evaluates distance-weighted nearest neighbors, boosted decision trees, random forests, artificial neural networks, support vector machines, and a heterogeneous multiple-model predictor. The unsupervised study develops general and case-based soil classifications using kernel k-means and kernel x-means, followed by neural-network models designed to reproduce the derived classes.

The dissertation uses a broad CPT dataset assembled from 179 soundings and serves as the common research foundation for several later publications and public code repositories.

Research lineage and related implementations

The repositories below preserve and document different branches of the dissertation's machine-learning work. They should be cited together with the corresponding journal or conference publication when used scientifically.

KNN

Distance-based feature analysis and DWKNN lineage. Companion to the 2019 Soils and Rocks article.

Trees

Recovered decision-tree and random-forest predecessor implementation used in the supervised research line.

ANN_SVM

Recovered ANN/SVM implementation lineage associated with the 2019 CILAMCE comparison and later supervised work.

MMP

Multiple-model predictor and supervised classification lineage associated with the 2021 Soils and Rocks article.

CPT-Kernel-Clustering

Kernel k-means, kernel x-means, and ANN reproduction of derived soil classes. Companion to the 2020 Soils and Rocks article.

Related peer-reviewed publications

Soil Classification System from Cone Penetration Test Data Applying Distance-Based Machine Learning Algorithms · Soils and Rocks, 2019.

Application of kernel k-means and kernel x-means clustering to obtain soil classes from cone penetration test data · Soils and Rocks, 2020.

A multiple model machine learning approach for soil classification from cone penetration test data · Soils and Rocks, 2021.

Citation

CARVALHO, Lucas Orbolato. Application of Machine Learning Techniques for Soil Classification from CPTU. 2018. 101 f. Master's dissertation (Aeronautical Infrastructure Engineering, Airport Infrastructure) - Instituto Tecnológico de Aeronáutica, São José dos Campos, Brazil, 2018. Advisor: Dimas Betioli Ribeiro.

BibTeX

@mastersthesis{Carvalho2018CPTU, author = {Carvalho, Lucas Orbolato}, title = {Application of Machine Learning Techniques for Soil Classification from CPTU}, school = {Instituto Tecnológico de Aeronáutica}, address = {São José dos Campos, Brazil}, year = {2018}, type = {Master's dissertation} }