Overview

When you click Run Prediction in Step 4: Select Input Features, DORA 2.0 executes its prospectivity modeling workflow using the inputs defined during setup.

During this process, DORA 2.0 prepares your data, trains and validates a machine learning model, and generates a Prediction Map.

The Prediction Map includes:

Surface VPS
Sub-surface VPS
Target Groups (generated as a final, optional step after the Prediction Map is complete)

The following sections explain what happens behind the scenes as DORA 2.0 processes your data and generates these results.

At a Glance: Prediction Workflow

When a prediction runs, DORA 2.0 processes your data through several stages:

Data Preparation – Selected datasets are standardized and aligned within the AOI
Data Fusion and Embedding – A transformer-based model analyzes relationships across datasets and converts them into embeddings
Model Training and Validation – The model learns from your Learning Data and is evaluated for performance
Prediction Outputs – The trained model is applied across the AOI to generate VPS results

Behind the Scenes: How DORA 2.0 Generates Predictions

Data Preparation

First, DORA 2.0 prepares all selected inputs for modelling, which includes:

Standardizing spatial data within the AOI
Aligning selected Input Features
Structuring Learning Data
Ensuring all datasets are compatible

All inputs are converted into a consistent format so they can be analyzed together.

Data Fusion and Embedding

Next, DORA 2.0 uses a transformer-based model to analyze relationships across multiple geological datasets.

These datasets are then converted into embeddings. Embeddings capture the combined signal from multiple data layers, allowing the model to represent spatial and geological context in a machine-readable format.

This enables DORA 2.0 to identify patterns associated with your selected Deposit Type and apply them across the AOI.

Model Training and Validation

Once the data has been prepared and embedded, DORA 2.0 trains a classification model using your Learning Data.

Learning Data includes labeled examples of mineralized and non-mineralized locations. The model learns the spatial patterns associated with these examples and applies that knowledge across the AOI.

To ensure reliable predictions, DORA 2.0 organizes learning points into spatial train–test clusters, grouping nearby points together geographically.

During training, the model trains on clusters from one area → A different cluster is held out for testing → The process rotates through clusters to evaluate performance.

This approach reduces spatial bias and helps ensure the model can generalize across the AOI.

Generating Prospectivity Outputs

Once the model has been trained and validated, DORA 2.0 applies it across the AOI to generate prospectivity outputs and performance graphs:

Surface VPS – Highlights prospectivity patterns at surface level across the AOI
Sub-surface VPS – Evaluates prospectivity below surface using depth-based information
Target Groups – Identify and organize clusters of high prospectivity (Target Groups can be created and adjusted as a final, optional step in the Prediction Map workflow)
Prediction Results – Display how well the model performed. Results include Feature Importance labels, prediction accuracy, and depth accuracy

Together, these outputs form the basis of the Prediction Map.

What Is a Prediction Map?

A Prediction Map is the final output generated by DORA 2.0..

It visualizes where the model identifies stronger prospectivity for your target commodity within the Area of Interest (AOI).

This is represented using the VPS, which reflects how closely different areas resemble the patterns in your Learning Data and selected Input Features.

The Prediction Map combines multiple outputs into a single view, helping you interpret results and identify potential exploration targets.