Publication
Overview
The study evaluates k-nearest neighbors and a Gaussian distance-weighted nearest-neighbor formulation for reproducing CPT-based soil classification systems across multiple raw and normalized feature combinations. The public repository preserves the historical datasets and archived workspaces while providing a cleaned, auditable implementation with validation-based model selection, deterministic rerun controls, automated tests, and explicit legacy behavior.
Reproducibility
The companion repository documents the 132-configuration experiment grid, preprocessing and class-balancing rules, historical implementation issues, cleaned paper-faithful behavior, and the preserved legacy artifacts.
BibTeX
@article{Carvalho2019CPTKNN,
author = {Carvalho, Lucas Orbolato and Ribeiro, Dimas Betioli},
title = {Soil Classification System from Cone Penetration Test Data Applying Distance-Based Machine Learning Algorithms},
journal = {Soils and Rocks},
volume = {42},
number = {2},
pages = {167--178},
year = {2019},
doi = {10.28927/SR.422167}
}