> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.umnai.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.umnai.com/_mcp/server.

# Feature and module attributions

Feature attribution describes contributions assigned to individual features. Module attribution describes contributions from learned components, including feature interactions. Both views explain the same submitted observation.

## Before you begin

Execute a [local explanation](/ui-guides/explanations/local-explanations) and keep its deployment and input record selected. Attribution tabs use that query's results.

## Inspect feature and module attributions

After executing a query, use the attribution tabs to see how the model’s components contribute to that result.

1. Select **Feature Attribution**.

2. Use **Attribution Direction** to show **All**, **Positive**, or **Negative** contributions.

3. Use **Feature Type** to narrow the view when needed.

4. Enable **Show Input Values** to see the values alongside the explanation.

5. Select a bar to highlight a feature. Use **Clear** to clear the selection.

6. Review the **Feature Attribution Table** below the plot. Its columns include Feature, Value, Attribution, Type, and Importance.

## Interpret the result

In the example, **relationship** has the largest visible negative feature contribution. This describes that query; feature contributions can differ for other records.

Switch to **Module Attribution** to inspect contributions at the module level. Use the direction filter and select a bar to highlight a module. A module can represent a single feature or a feature interaction, such as **educational_45_num x relationship**.

Feature and module views are different ways to inspect the explanation. Do not add their values together across views as if they were independent contributions.

**Interpretation:** positive and negative attribution describe contribution in the model’s explanation scale. Attribution does not, by itself, establish a causal relationship or justify changing a real-world attribute.

![Feature attribution filters and the plotted contributions for the query.](/_fern-files/_img/e0b52f3b21883372694004b8345da697398b74bc0a6a488502c5e546b4b9dce7.webp)

![The visible portion of the module attribution plot for the same query.](/_fern-files/_img/a477e766641e9dbdbbbe58353d1b19d7cadd04818498696669bce5cf38095926.webp)

## Choose display options

| Control                   | Use it to                                                                         |
| ------------------------- | --------------------------------------------------------------------------------- |
| Attribution Direction     | Focus on positive or negative contributions, or select All for the complete view. |
| Feature Type              | Restrict feature attribution to the available feature types.                      |
| Show Input Values         | Read the record's values alongside their contributions.                           |
| Bar selection and Clear   | Highlight a feature or module, then remove that selection.                        |
| Feature Attribution Table | Compare labels, input values and numeric attribution values below the plot.       |

A filtered plot omits contributions outside the selected filter. Restore **All** and clear feature-type filters before judging the complete explanation. The largest contribution in this record is not necessarily the most important feature across the population.

## Next steps

Continue to [Model structure and decision rules](/ui-guides/explanations/decision-rules).