Visualizing SHAP explainability

Explainable AI in Python

Fouad Trad

Machine Learning Engineer

Dataset

age	gender	bmi	children	smoker	charges
19	0	27.900	0	1	16884.92
18	1	33.770	1	0	1725.55
28	1	33.000	3	0	4449.46
33	1	22.705	0	0	21984.47
32	1	28.880	0	0	3866.85

model: random forest regressor to predict charges

import shap
explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(X)

shap.summary_plot(shap_values, X, plot_type="bar")

Feature importance plot showing that smoking status is the most important feature.

Shows SHAP values distribution
Highlights direction and magnitude of each feature on prediction
- Red color → high feature value
- Blue color → low feature value
- SHAP value > 0 → increases outcome
- SHAP value < 0 → decreases outcome

Beeswarm plot showing shap value distribution for each feature, with colors indicating feature values.

Shows SHAP values distribution
Highlights direction and magnitude of each feature on prediction
- Red color → high feature value
- Blue color → low feature value
- SHAP value > 0 → increases outcome
- SHAP value < 0 → decreases outcome

Same beeswarm plot highlighting that high feature values of the 'smoker' feature have positive SHAP values.

Shows SHAP values distribution
Highlights direction and magnitude of each feature on prediction
- Red color → high feature value
- Blue color → low feature value
- SHAP value > 0 → increases outcome
- SHAP value < 0 → decreases outcome

shap.summary_plot(shap_values, X, 
                  plot_type="dot")

Same beeswarm plot highlighting that low feature values of the 'smoker' feature have negative SHAP values.

A generic partial dependence plot showing the relation between a feature and an outcome as a decreasing function.

For each sample:
- Vary value of selected feature
- Hold other features constant
- Predict outcome
Average results from all samples

shap.partial_dependence_plot("age", 
                             model.predict, 
                             X)

Partial dependence plot for the age feature showing the when age increases, the predicted value increases.

For each sample:
- Vary value of selected feature
- Hold other features constant
- Predict outcome
Average results from all samples

shap.partial_dependence_plot("age", 
                             model.predict, 
                             X)

Same partial dependence plot highlighting the distribution of the age values in the dataset.

Explainable AI in Python