Overall attrition stands at 16.1% across 1470 employees (logistic AUC = 0.848). The single strongest logistic predictor is Job Role: Sales Representative (OR = 6.23). XGBoost ranks Monthly Income as the highest-gain feature.

Attrition rates are broken down across department, overtime status, business travel frequency, and job level. The highest-attrition segment is Overtime — Yes — at 30.5% (127 attritions / 416 employees). Groups with fewer than 5 employees are excluded.

Sales Representative has the highest attrition rate at 39.8%, while Research Director has the lowest at 2.5%. Across all 9 roles shown, the median attrition rate is 16.1%. Roles with fewer than 5 employees are excluded.

Showing the top 10 predictors by absolute log-odds magnitude. Job Role: Sales Representative has the highest odds ratio (OR = 6.23, 95% CI: 0.66–69.07), meaning employees with this attribute are 6.2x more likely to attrite. 7 of 10 shown predictors are associated with increased attrition (OR > 1). Predictors with extreme or infinite ORs (separation) are excluded.

Full logistic regression results for all 28 predictors (after dummy coding). 17 predictors are statistically significant at p < 0.05. Odds ratios and CIs are exponentiated from log-odds; NA/Inf values indicate near-complete separation and should be interpreted with caution.

XGBoost gain-based feature importance (proxy for mean absolute SHAP values) across the top 10 drivers. Monthly Income contributes the most gain to the model's attrition predictions, suggesting it provides the largest non-linear discrimination between employees who leave and those who stay. Unlike logistic odds ratios, these ranks capture interaction effects.

The overtime × department interaction reveals where workload and organisational context combine to elevate attrition risk. The highest-risk cell is Yes employees in Sales at 37.5% attrition. Cells with fewer than 5 employees are excluded from this view.

The Kaplan-Meier survival curve shows employee retention probability over tenure. After 33 years, an estimated 51.0% of employees remain. By year 5, 87.3% of employees are still with the company. Steep early drops indicate the first few years carry the highest attrition risk.

The ROC curve summarises the logistic regression model's ability to distinguish employees who will attrite from those who will not. AUC = 0.848 indicates good discriminative performance. A diagonal line would represent random guessing (AUC = 0.5); the further the curve bows toward the top-left, the better the model.