Publication
Overview
The study applies Gaussian kernel k-means and kernel X-means to derive soil classes directly from measured cone penetration test variables. The Full analysis evaluates candidate cluster counts on 130,966 CPT observations, while the Specific analysis uses X-means-style BIC splitting on a 2,847-observation subset. Artificial neural networks are then used to quantify how consistently the resulting cluster labels can be learned from the CPT measurements.
Reproducibility
The companion repository preserves the recovered author-origin development source by checksum, provides cleaned deterministic runners for kernel clustering and ANN evaluation, documents the historical preprocessing and implementation limitations, and deliberately excludes datasets and third-party X-means source for which redistribution rights are not established.