95 lines
3.5 KiB
JSON
95 lines
3.5 KiB
JSON
{
|
|
"_comment": "Example JSON output for a model explanation. This is a template for the model-explainability-tool plugin.",
|
|
"model_id": "model_v3.2",
|
|
"model_type": "Classification",
|
|
"dataset_used": "customer_churn_dataset.csv",
|
|
"explanation_type": "SHAP",
|
|
"explanation_timestamp": "2024-01-26T10:30:00Z",
|
|
"global_explanation": {
|
|
"_comment": "Global feature importance ranking.",
|
|
"feature_importance": [
|
|
{
|
|
"feature": "contract_length",
|
|
"importance_score": 0.35,
|
|
"description": "Length of the customer's contract (e.g., monthly, yearly)."
|
|
},
|
|
{
|
|
"feature": "monthly_charges",
|
|
"importance_score": 0.28,
|
|
"description": "The customer's monthly bill amount."
|
|
},
|
|
{
|
|
"feature": "total_charges",
|
|
"importance_score": 0.22,
|
|
"description": "Total amount the customer has paid."
|
|
},
|
|
{
|
|
"feature": "internet_service",
|
|
"importance_score": 0.10,
|
|
"description": "Type of internet service the customer has (e.g., DSL, Fiber optic)."
|
|
},
|
|
{
|
|
"feature": "online_security",
|
|
"importance_score": 0.05,
|
|
"description": "Whether the customer has online security."
|
|
}
|
|
],
|
|
"summary": "The model's predictions are most influenced by contract length, monthly charges, and total charges. Internet service and online security have a smaller, but still significant, impact."
|
|
},
|
|
"local_explanation": {
|
|
"_comment": "Explanation for a specific instance/prediction.",
|
|
"instance_id": "customer_123",
|
|
"predicted_class": "Churn",
|
|
"prediction_probability": 0.85,
|
|
"feature_contributions": [
|
|
{
|
|
"feature": "contract_length",
|
|
"contribution": -0.40,
|
|
"value": "Month-to-month",
|
|
"reason": "Month-to-month contracts are highly correlated with churn."
|
|
},
|
|
{
|
|
"feature": "monthly_charges",
|
|
"contribution": 0.25,
|
|
"value": 75.50,
|
|
"reason": "Higher monthly charges increase the likelihood of churn."
|
|
},
|
|
{
|
|
"feature": "total_charges",
|
|
"contribution": -0.10,
|
|
"value": 200.00,
|
|
"reason": "Relatively low total charges suggest the customer is new and more likely to churn."
|
|
},
|
|
{
|
|
"feature": "internet_service",
|
|
"contribution": 0.05,
|
|
"value": "Fiber optic",
|
|
"reason": "Fiber optic service is associated with higher churn rates in this dataset."
|
|
},
|
|
{
|
|
"feature": "online_security",
|
|
"contribution": -0.02,
|
|
"value": "No",
|
|
"reason": "Lack of online security slightly increases churn risk."
|
|
}
|
|
],
|
|
"summary": "This customer is predicted to churn primarily due to their month-to-month contract and high monthly charges. The relatively low total charges also contribute to the prediction."
|
|
},
|
|
"fairness_metrics": {
|
|
"_comment": "Metrics for assessing fairness across different groups.",
|
|
"protected_attribute": "gender",
|
|
"metric": "Disparate Impact",
|
|
"value": 0.95,
|
|
"threshold": 0.8,
|
|
"status": "Acceptable",
|
|
"summary": "The model exhibits acceptable disparate impact across genders, as the value (0.95) is above the threshold (0.8)."
|
|
},
|
|
"data_bias_detection": {
|
|
"_comment": "Results of data bias detection.",
|
|
"potential_bias": "Unequal representation of geographic regions in the training data.",
|
|
"recommendations": [
|
|
"Collect more data from underrepresented regions.",
|
|
"Use re-weighting techniques to balance the data."
|
|
]
|
|
}
|
|
} |