In this article, you’ll learn how to select and configure Input Features within your experiment in DORA. In this step, you choose the geoscience data you want to include in the experiment. Additionally, you can adjust features, visualization options, and correlation thresholds, allowing you to highlight specific data layers and emphasize the details most relevant to your experiment.

Input Features are the exploration data grids used in the model. These come from the datasets compiled for the project (this includes your project data) and are synchronized with the 3D Layers List. You can select up to 64 feature layers for an experiment.

As a best practice, begin by selecting input features that geologists consider most relevant, specifically features known to correlate with mineralization or supported by past exploration success. These are often the most geologically meaningful and will anchor your model in known exploration logic.
​

Additionally, make sure your selected features match the coverage of your AOI to ensure full spatial alignment. Higher resolution doesn’t always mean higher accuracy, especially if your input raster only covers part of the AOI. DORA will extrapolate the missing areas, but this can reduce prediction reliability. It's often better to use a slightly weaker feature that covers 100% of your AOI than a stronger one that only covers 50%.

Using input features helps geoscientists align the model with geological reality. Layers such as faults, soil anomalies, and geophysics ensure that predictions are connected to an exploration strategy. By testing different combinations (e.g., geophysics only or soils only), geoscientists can explore new hypotheses and gain insight into mineralization controls.

Input features are the variables the model learns from, and their quality and relevance directly affect accuracy. Interpolated data fills gaps in coverage, while point data is rasterized to create consistent grids across the AOI. The correlation threshold reduces duplication, preventing the model from being skewed by redundant features. Even lower-quality ("lowQC") features are retained for completeness, but are flagged so users are aware of their reliability.

Start with High-Confidence Features

Use your most reliable data (like geophysical layers) to reduce model uncertainty. Avoid overloading the model, as too many features can dilute patterns and limit target results. Begin with a few strong layers, then add others gradually to explore new ideas.

Experiment and Interpret

Running multiple experiments with different combinations of features is encouraged. If you're consistently seeing similar target areas even as you vary layers, this consistency can help you build geological confidence in those predictions.

Use Low-QC Features Cautiously

Avoid using low-quality (Low-QC) features unless you have a strong reason to include them. While they may seem helpful in early-stage exploration, they introduce more uncertainty and can reduce the model’s reliability. Focus first on high-confidence data sources before experimenting with lower-quality layers.

Filling Data Gaps Through Extrapolation

When your learning data has gaps, DORA uses complete, continuous input layers (like magnetics or remote sensing) to extrapolate patterns from well-sampled areas into unsampled ones. This helps create a full prediction surface and improves confidence in areas without direct samples.

Final Trade-off Reminder

The best features aren’t just the ones you trust most — they also need strong coverage and appropriate resolution. A feature that covers only part of your AOI or mismatches other layers in resolution may hurt model performance.

Remember to balance data quality, coverage, and relevance for the best results.

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