Studio overview
Understand the workspace and connected project model.
Explore →GEOLAYERS TUTORIALBuild
Build
a GeoML Prediction Workflow
A practical step-by-step guide to build a geoml prediction workflow in the GeoLayers platform.
Browser-based GISGeoAI + spatial analyticsSouth African context
LEARNING GOAL
Build a GeoML Prediction Workflow
Train a simple predictive model from project attributes and evaluate it before mapping the predictions.
1. Define the targetChoose whether the task is regression, classification, clustering or suitability. For supervised models, select a target field that represents the outcome you want to explain or predict.
2. Choose predictorsSelect fields that are available at prediction time and are logically related to the target. Avoid leaking the answer into the predictor set.
3. Validate spatiallyStart with a holdout and use spatial-block validation when nearby observations could make random train/test splits overly optimistic.
4. Read the metricsFor regression inspect R², RMSE and MAE; for classification inspect accuracy, Macro F1 and the confusion matrix.
5. Publish the predictionWrite predictions/probabilities to a project layer, map the result and compare unexpected areas back to the source evidence.
GOOD PRACTICE
What to check before you finish
A tutorial is complete when the user can explain the result, not just reproduce the clicks.
- Keep source data and assumptions visible in the project.
- Use clear layer and field names that will still make sense when the project is reopened.
- Inspect attributes and metrics before interpreting a visual result.
- Save or publish outputs in a form that can be reviewed by another person.
CHECK YOUR WORK
Common quality checks before you move on
A few checks can prevent a technically correct workflow from becoming a misleading output.
- If training performance is high but spatial holdout performance collapses, investigate spatial leakage or overfitting.
- Check whether predictors would actually be available when the model is applied to new features.
- Do not interpret a prediction map as certainty; review diagnostics and unexpected locations.