Prediction Models

Model options exposed by the prediction dialog and how to review them.

Model Selection

The prediction dialog shows model options that are compatible with the selected target and available scene data. The goal is to make prediction runs comparable: choose a target, select a model, review the relevant parameters, then inspect the resulting layer against the source data.

Displayed Model Types

Ordinary Kriging
Prediction dialog option

A geostatistical estimator exposed for supported assay and lithology targets. Review the variogram choice before running it.

Inverse Distance Weighting
Prediction dialog option

A distance-weighted estimator for supported continuous assay targets. The dialog exposes neighbor-count controls where applicable.

Nearest Neighbors
Prediction dialog option

A local estimator that uses nearby samples. It is useful as a simple comparison run when the target supports it.

Radial Basis Function
Prediction dialog option

An interpolation option for supported assay targets. Review the selected kernel before comparing it against other runs.

Tree-Based ML
Prediction dialog option

A machine-learning model option for supported targets. Treat it as one comparable run in the prediction job history.

Deep Learning
Prediction dialog option

A model option for supported targets. Use the same input review and output comparison process you use for every prediction run.

Controls to Check Before Running

  • Target type: assay or lithology when the selected 3D scene has the necessary inputs.
  • Assay element and measurement method for assay predictions.
  • Output format, such as surfaces or block model, depending on the selected target.
  • Model-specific options such as variogram, neighbor count, kernel, cube size, and smoothing controls.
  • Runtime estimate and job status, shown before and after submission.

Comparing Results

Completed jobs appear in Past Predictions with the model type, target, status, date, and any displayed model parameters. Use that history to compare runs. A different model or parameter set should be judged by how it lines up with the original drill data and geology, not by the model name alone.