Calculated explanation
Local SHAP force plot
Automatic interpretation
Validation messages
Detailed feature contribution table
| Rank | Feature | Value | SHAP | |SHAP| | Share | Direction | Cumulative output | Status | Interpretation |
|---|
Global feature importance across observations
Observation summary
| Observation | Prediction | Top feature |
|---|
Formula used
Prediction = Base value + Σ SHAP valueᵢ Reconstruction error = Entered prediction − Calculated prediction Contribution shareᵢ = |SHAP valueᵢ| ÷ Σ|SHAP values| × 100%
A SHAP force plot begins at the model baseline. Each feature moves the output upward or downward. The final point is the reconstructed prediction value.
Positive contributions increase the explained model output. Negative contributions reduce the explained model output. SHAP values describe model behavior, not proven causation.
How to use the calculator
- Select the model task and explained output scale.
- Enter the baseline and optional model prediction.
- Add feature names, observed values, and signed SHAP values.
- Choose sorting, thresholds, labels, and comparison settings.
- Calculate, validate totals, inspect explanations, and export results.
Example data format
| observation | feature | value | shap | base | prediction |
|---|---|---|---|---|---|
| customer-001 | Contract type | Month-to-month | 0.18 | 0.42 | 0.71 |
| customer-001 | Online security | Enabled | -0.06 | 0.42 | 0.71 |
Frequently asked questions
What does a SHAP force plot show?
It shows how features move one prediction from its baseline. Positive effects raise the output, while negative effects reduce it. The ending position represents the explained model prediction.
What is the baseline value?
The baseline is the model’s expected output before local features. It often reflects an average prediction over background data. Every displayed contribution starts from this reference value.
Can I use probability SHAP values?
Yes, when your explainer returns values on probability scale. Their sum should reconstruct the selected class probability. Confirm the explainer output scale before interpreting results.
Why does the reconstructed prediction differ?
The difference may come from rounding or mismatched output scales. Missing features can also create a visible reconstruction error. Check every contribution and the baseline carefully first.
What do positive and negative SHAP values mean?
Positive values push the explained output above its baseline. Negative values pull the explained output below its baseline. Their meaning depends on the selected class and scale.
Does a SHAP value prove causation?
No, SHAP explains model behavior using the supplied inputs. Correlated features may share or redistribute apparent contribution. Use domain evidence before making causal conclusions.
Can I compare observations?
Yes, enable comparison and select another imported observation. The calculator displays both local force plots side by side. Compare direction, magnitude, and final prediction carefully.
How are small features grouped?
Features outside thresholds or display limits can become Other. Their signed contributions remain included in the reconstructed prediction. Grouping improves readability without changing total contribution.
Where is my project data stored?
Named projects are stored inside your current browser storage. They remain local unless you export and share them. Clearing browser data can remove saved projects permanently.