Semi-Supervised Ensemble Learning

Weighted-ensemble regression over base learners supported by 'caret' (Kuhn (2008) ), with cross-validated hyperparameter selection, out-of-fold diagnostics, and signed residual- offset estimates. Multi-response problems use iterative input-space expansion related to Spyromitros-Xioufis et al. (2016) , with Jacobi or 'Gauss-Seidel' sweeps, package-defined companion gates and per-response iteration stitching. A package-defined pseudo-label stage promotes prediction rows by a cross-model and cross-dataset range ratio and accepts rounds with an out-of-fold squared-correlation gauge.

ssel

Semi-Supervised Ensemble Learning

ssel provides three related method layers for tabular regression:

weighted ensembles over caret-supported learners, with cross-validated tuning, out-of-fold diagnostics, signed residual-offset estimates, and two explicit response/dataset selection policies;
iterative multi-response input-space expansion with Jacobi or asymmetric Gauss--Seidel companion updates, package-defined fixed or shadow gates, and per-response iteration stitching;
package-defined pseudo-label promotion based on a cross-model and cross-dataset range ratio, with an out-of-fold squared-correlation gauge for accepting or reverting candidate rounds.

The default supervised learners are random forest (ranger), gradient boosting (gbm), and multivariate adaptive regression splines (earth). Any compatible caret method can be supplied through the documented method arguments.

Installation

ssel requires R >= 4.1.0.

# Released version
install.packages("ssel")

Start here

The quickstart builds a minimal in-memory dataset and introduces the public entry points:

vignette("ssel-quickstart", package = "ssel")

The method articles define the mathematics, package policies, API mapping, and limitations of each helper family:

vignette("ensemble-theory",  package = "ssel")
vignette("chain-regression", package = "ssel")
vignette("semi-supervised",  package = "ssel")

Pipeline entry points compose the lower-level helpers and write their results to caller-supplied directories. Their exact input schemas, side effects, and failure conditions are documented in the function reference:

?modelPipeline
?chainPipeline
?semiSupervisedPipeline

Rendered documentation: https://averriK.github.io/ssel/.

Public API

Pipeline entry points: modelPipeline(), chainPipeline(), semiSupervisedPipeline(), postProcess().

Dataset assembly: buildDataset().

Training and prediction: trainModel(), trainRegressionModel(), predictModel(), aggregateResponses().

OOF reconstruction and diagnostics: oofEnsemble(), auditQuantiles(), auditOverfit(), detectOutliers().

Companion gates: extractChainImportance(), activeByImportance(), activeByShadow(), computeActiveByImportance().

Utilities: toNumeric(), which.nonnum(), removeOutliersIQR().

Full reference: help(package = "ssel") or https://averriK.github.io/ssel/reference/.

Citation

@Manual{VerriKozlowski_ssel,
  title  = {{ssel}: Semi-Supervised Ensemble Learning},
  author = {Verri Kozlowski, Alejandro},
  year   = {2026},
  note   = {R package version 0.3.1},
  url    = {https://averriK.github.io/ssel/}
}

License and contact

MIT; see LICENSE. The maintainer is Alejandro Verri Kozlowski (ORCID 0000-0002-8535-1170), Universidad de Buenos Aires, Facultad de Ingenieria. Contact information is listed in DESCRIPTION.