Overview
AI helps uncover patterns and predictions within large datasets that would take humans much longer to identify, if ever. In DORA’s context (supervised machine learning), the goal is to train models that learn real geological patterns without becoming too rigid or too vague. This balance is critical.
Overfitting happens when the model memorizes all of the data
Underfitting happens when the model learns too little.
Both reduce the effectiveness of AI predictions in mineral exploration.
What is Overfitting?
Overfitting occurs when a model learns the training data too well, including noise or irrelevant details. It may perform well on known data but fails to generalize to new or slightly different data.
In DORA, this means the model may only recognize mineralization patterns that closely match the training examples, missing other valid patterns outside of the learning points.
💡 Think of a model like a student studying for a test.
Overfitting = Memorized the practice test
The student memorized all the practice questions.
On test day, the questions are slightly different, and they struggle.
They ace the practice test, but perform poorly on anything new.
What is Underfitting?
Underfitting happens when a model is too simple and fails to learn important patterns in the data. In DORA, this leads to poor predictions on both training data and project-specific data, with predictions often appearing random or unreliable.
💡 Continuing the student example:
Underfitting = Studied too little
The student skimmed a few pages.
They don’t understand key concepts.
On test day, they can’t answer even basic questions.
Avoid Overfitting and Underfitting
Two key DORA outputs that will help you assess your model’s performance are the Prediction Accuracy gauge and Performance Breakdown matrix.
These tools help determine at a glance whether your model is overfitted or underfitted.
Example:
💡 The goal is: Just right (generalization)
The student studies the core concepts and practices various questions.
They grasp the ideas, not just the examples.
On test day, they can handle new questions confidently.
If your model is underfitted or overfitted, reach out to your DORA contact for help adjusting the model to improve results.
Learn More
Create a Prediction Map:
Still Have Questions?
Reach out to your dedicated DORA contact or email support@VRIFY.com for more information.