Highest accuracy · you own it

Your own machine-learning model, trained on your data.

We train a model on your area — your choice of method — using satellite imagery, the national geoscience dataset, and your own samples. Trained on your ground, it's the most accurate option when you have enough local data. And the model is yours to keep.

What goes into your model.

We pull every relevant layer for your area, line it up, and train on it — plus anything you bring.

Satellite

Satellite imagery

Alteration and surface signals from Sentinel, Landsat and ASTER, pulled through Google Earth Engine.

Geophysics

National geophysics

Gravity, magnetics and radiometrics — the layers that reveal buried structures.

Your data

Your samples

Your soil, rock-chip or drill assays, folded in to sharpen the model for your ground.

Our base

The national dataset

3.77M samples, mapped geology, occurrences — so your model learns from the whole continent, not just your patch.

Your choice of method.

Random forest, gradient boosting, or whichever approach suits your data — we'll recommend one, or run what you ask for. We tune it to your rock types and your commodity.

A model trained on local ground often scores higher than a general engine — in our own tests, a tuned model has beaten our standard approach on accuracy. When that's what your ground needs, a custom model is the way to go.

It's still honest. We test it the same hard way — hide the known deposits, then check it finds them — show you which factors drove it, and flag the ground where it's working outside what it has seen. And if the test says your data can't support a trustworthy model, we tell you straight — instead of shipping one that just memorised where the samples are.

Methods we run

Random forestrobust, fast, a strong default
Gradient boostingoften the highest accuracy
YDF decision forestsstrong with mixed data types
Your preferred approachtell us what you use

What you get back.

The same clear package as every MineDSS job — plus the model itself.

1

A map of ranked drill targets

Every promising spot on your ground, scored and ranked from best to worst.

2

The evidence behind every target

Host rock, geophysics, geochemistry, alteration — laid out so your team can check our work.

3

A trust flag on every target

Solid evidence, thin evidence, or a guess — so you know which targets to back and which to skip.

4

A drilling shortlist in priority order

Your best targets first, balancing how promising each one is against how sure we are.

5

A report you can hand over

Plain English for your board, your JV partner and your geologist — every line backed by the evidence.

The model is yours.

You keep the trained model and its outputs. Re-run it on new data whenever you like. Your data is never put into anyone else's model.

Choose a custom model when…

  • You have your own samples or drilling to train on.
  • Accuracy on your ground matters more than a step-by-step reasoning chain.
  • You want to own the model and re-run it yourself.

Need to defend every target line by line, or working with thin local data? mineBAG may fit better. We'll tell you which.

A model built for your ground.

Send us your area and your data — we'll scope a custom model and quote it up front.

Scope a pilot AOI