Feature Importance
SHAP feature importance · XGBoost model
Risk Score Distribution
Customer count by predicted churn probability bucket
Predicted vs Actual Churn by Segment
Churn rate % · grouped by policy segment
Model Confusion Matrix
Validation set · n = 98,240
True Positive
4,821
Correctly predicted churn
False Positive
912
Predicted churn, stayed
False Negative
1,240
Missed churners
True Negative
91,267
Correctly predicted stay
Overall Accuracy: 97.5% | Precision: 84.2% | Recall: 79.6%
High-Risk Customer Segments
Segments requiring immediate intervention
| Segment |
At-Risk Count |
Churn Prob |
Action |
| Auto Only |
5,821 |
9.1% |
Retention offer |
| Tenants |
4,201 |
11.4% |
Priority outreach |
| Age 18–25 |
2,890 |
12.8% |
Digital campaign |
| Low Tenure |
2,417 |
14.2% |
Onboarding review |
Churn Rate by Province
Annual churn % · current fiscal year