Machine Learning: Predict New Properties

This tutorial demonstrates how the results of the Train Model derived from Machine Learning (ML) can be used to predict new material properties by linear regression.

The Electronic Band Gap calculated in the training tutorial for Si/Ge-based materials is used as the example, though the approach works for many different target properties.

1. Pre-requisite: trained model

This tutorial assumes that an ML model in the workflow called "ml_predict" has already been trained to predict the band gap of Si/Ge-based materials, following the steps in the training tutorial.

2. Create the ML Predict job

A new "ML Predict" Job can be set up by following the general instructions for creating a new Job.

3. Select the trained model as workflow

The "ml_predict" workflow should be selected as the main Workflow for the job. This applies the trained model to predict properties of new materials similar to those used in training.

4. Select the target properties

The properties to be predicted are the target properties selected under the unit editor of the "input" unit of the "ml_predict" workflow, in the "Targets" section.

5. Submit the job

The "ML Predict" job can be executed after configuration in Job Designer.

6. View the results

The predicted properties are available under the Results tab of Job Viewer.

7. Video walkthrough

The animation below demonstrates predicting the band gap of Si₄Ge₁₂ using the model trained in the training tutorial. The ML-predicted direct and indirect band gaps (0.525 and 0.490 eV) are in good agreement with the DFT-calculated values (0.517 and 0.441 eV).