SHAP Summary Plot Calculator

Calculated SHAP Results

Sample data is ready. Select Calculate Plot to generate results.

Most influential feature

—

Strongest positive contributor

—

Strongest negative contributor

—

Total absolute contribution

—

Samples used

—

Features displayed

—

Missing values

—

Selected class

All

Automatic interpretation

Generate the plot to receive a concise interpretation.

Feature Statistics

Rank	Feature	Mean \|SHAP\|	Mean SHAP	Median	Minimum	Maximum	Std. deviation	Positive	Negative	Zero	Contribution	Value correlation
No calculated statistics yet.

1. SHAP Data Input

Delimiter

Combined paired CSV

Use columns such as Age__shap and Age__value.

Optional base value

SHAP value matrix

Include feature headers. The first sample identifier column is optional.

Feature value matrix

Feature columns should match the SHAP matrix.

Class or group labels

Actual labels

Prediction labels

2. Plot and Ranking Options

Plot type

Feature ranking

Ranking direction

Maximum displayed features

Maximum displayed samples

Decimal precision

Point size

Current: 8 pixels

Point opacity

Current: 0.75

Horizontal row jitter

Current: 0.28

Feature row spacing

Current: 1.0

SHAP axis range

Custom minimum

Custom maximum

Chart title

Horizontal axis label

Chart height

Plot font size

Legend position

Number notation

Show zero reference line

Show gridlines

Show overlapping-point counts

Highlight extreme impacts

Add values to feature labels

3. Colour, Class, and Filtering Controls

Point colour mode

Low colour

High colour

Fixed colour

Missing colour

Class or group filter

Label filter field

Required label value

Impact direction filter

Minimum absolute SHAP magnitude

Feature value minimum

Feature value maximum

Search feature names

Specific features

Select one or more features. Leave all selected to include everything.

Reverse low-to-high colour scale

Show colour legend

Exclude rows containing missing values

Skip invalid rows when possible

Average importance across classes

4. Optional Dataset or Model Comparison

Comparison view

Comparison dataset label

Comparison paired CSV

Comparison SHAP matrix

Feature names should overlap the primary dataset.

5. Calculate, Save, and Export

Formula Used

Global feature importance: Mean absolute SHAP importance equals the average absolute SHAP value for each feature.

Importance(j) = (1 / n) × Σ |SHAP(i, j)|

Signed average: Mean SHAP preserves impact direction. Positive values increase the explained model output.

SignedMean(j) = (1 / n) × Σ SHAP(i, j)

Contribution percentage: Each feature importance is divided by total feature importance, then multiplied by one hundred.

How to Use

Paste SHAP values with feature names in the first matrix.
Add matching feature values for meaningful point colours.
Optionally enter classes, actual labels, and predictions.
Select ranking, filters, colours, and a plot type.
Choose Calculate Plot to validate and analyse everything.
Inspect tooltips, statistics, warnings, and automatic interpretation.
Export the chart, report, cleaned data, or statistics.

Example Data Format

sample	Age	Income	Credit Score
C-001	0.31	-0.18	-0.24
C-002	-0.12	-0.22	-0.19

A combined upload may use paired columns. Example names include Age__shap and Age__value.

Interpretation Guide

Features near the top have greater average influence. Wider horizontal spreads indicate more variable model effects.

Points right of zero raise the explained output. Points left of zero lower the explained output.

Feature-value colours reveal directional relationships. Similar colours on both sides may indicate interactions or nonlinear effects.

SHAP explanations describe model behaviour, not proven causation. Validate findings with domain evidence and controlled analysis.

Frequently Asked Questions

What does a SHAP summary plot show?

It combines global feature importance with individual observation impacts. The horizontal position shows impact direction and magnitude.

Why use mean absolute SHAP values?

Absolute values prevent positive and negative effects from cancelling. Their average creates a useful global importance score.

What do the colours represent?

Colours usually represent original feature values. Low and high colours help reveal directional feature relationships.

Can regression SHAP values be used?

Yes. Regression values explain movement from the model base value toward each prediction.

How are multiclass results handled?

Enter class labels and filter a class. The stacked view compares mean importance across available groups.

What happens with missing values?

Missing values receive a separate colour by default. You may exclude incomplete rows before calculating.

Why must matrix dimensions match?

Every SHAP value needs the corresponding feature value. Matching rows and columns preserve observation alignment.

Can two model versions be compared?

Yes. Add a comparison SHAP matrix and select importance, rank, or percentage change.

Does high SHAP importance prove causation?

No. SHAP values explain model predictions. They do not establish causal relationships in the underlying system.