Skip to main content

How DORA Scores and Ranks Input Features

Learn how DORA automatically evaluates and ranks raster inputs in Step 4: Select Input Features, including how composite scores are calculated, what each criterion measures, and how AOI coverage is displayed separately.

Overview

In Step 4: Select Input Features, DORA automatically evaluates raster inputs before modelling.

Each raster receives a composite score from 0–100 based on four equally weighted criteria. The final score is the average of these four scores.

AOI Coverage is evaluated separately and is not included in the composite score. It shows the percentage of your selected AOI that the raster covers.

Rasters are ranked by their composite score, and the top layers (up to 64) are automatically included in the Recommended Feature selection. You can review and adjust this selection before running the model.

Because the evaluation considers multiple factors, a raster’s score reflects its geological relevance, redundancy with other inputs, and compatibility with the modelling configuration.

Note: If an input feature has been heavily interpolated or processed through a Data Augmentation module, its score may not fully reflect the quality of the underlying data. Where possible, use the original raster alongside the processed version and compare scores. The raw raster score is the more reliable signal.

How DORA and ranks input features

How DORA Evaluates Input Features

DORA evaluates each raster using four criteria that measure geological relevance, redundancy, and compatibility with the modelling configuration.

The scores are designed to identify hidden issues such as instability, aliasing, scale inconsistencies, noisy spatial frequencies, sampling artifacts, and weak foundational training characteristics that are often difficult or impossible to reliably detect through visual inspection alone.

1. Mineralization Relationship

  • What it measures:

    • How strongly the raster distinguishes mineralized from unmineralized areas, based on its relationship to the rasterized Learning Data. Because this score is calculated from rasterized Learning Points rather than raw sample locations, it is influenced by the resolution of the model. (Step 2: Set Up Learning Data)

  • How rasters are scored:

    • Stronger relationships with mineralization = higher scores.

    • Little or no relationship to the target mineral = lower scores.

  • Why it matters:

    • Features that are strongly associated with mineralization contribute more meaningful predictive signals.

2. Model Alignment

  • What it measures:

  • How rasters are scored:

    • Similar to data the foundation model was trained on = higher scores.

    • Entirely new or unfamiliar to the model = lower scores.

  • Why it matters:

    • Models perform best when applied to data types similar to their training data.

3. Raster Redundancy (Collinearity)

  • What it measures:

    • The degree of similarity between a raster and other selected rasters.

  • How rasters are scored:

    • If two rasters contain highly similar information, one will be penalized.

    • DORA prioritizes the higher-scoring raster to reduce duplication.

  • Why it matters:

    • Highly correlated layers do not add new information and can reduce model efficiency.

4. Frequency and Sampling Compatibility

  • What it measures:

    • Whether the raster’s spatial resolution and sampling characteristics are appropriate for the AOI and modelling scale.

    • This criterion combines three related checks that evaluate how well a raster's spatial resolution and sampling align with your AOI and modelling scale. The score displayed is the composite of these three sub-scores.

  • How rasters are scored:

    • Resolution Compatibility

      • Too coarse relative to the AOI = lower scores (including oversampled rasters that have been over-averaged and lose important signal)

      • Too fine relative to the AOI = lower scores if it introduces noise or does not add meaningful detail (including undersampled rasters that have been over-interpolated without adding new information)

    • Spatial Scale Alignment

      • Spatial patterns are misaligned with the scale of the AOI = lower scores

      • If the AOI is too small to capture the raster’s dominant spatial patterns = lower scores

    • Sampling Consistency

      • Under- or over-sampled relative to the modelling resolution = lower scores

      • Inconsistent sampling across the raster = lower scores

  • Why it matters:

    • Rasters must be spatially compatible with the modelling scale to provide reliable predictive signals. Poor alignment in resolution, sampling, or spatial scale can distort or obscure meaningful geological patterns.

Click the composite score to open the expanded feature scores:

How to read the expanded feature scores:

Prediction Score breakdown

AOI Coverage

AOI Coverage is displayed separately from the composite score. It shows the percentage of your selected AOI that the raster covers.

High coverage means the raster data extends across most or all of your AOI.

Low coverage means parts of your AOI have missing data, which may reduce model reliability or introduce bias.

How DORA Uses the Score

Rasters are ranked from highest to lowest based on their composite score.

Features are colour-coded to help you quickly assess their suitability:

  • Green: 81–100

  • Yellow: 51–80

  • Grey: 0–50

The top-ranked layers (up to 64) are automatically included in the Recommended Feature selection.

This provides a strong, data-driven starting point, but the selection should be reviewed and adjusted based on your geological understanding.

Read Recommended Workflow: Input Feature Selection for more information about adjusting features.

Learn More

Still Have Questions?

Reach out to your dedicated DORA contact or email support@vrify.com for more information.

Did this answer your question?