Executive Summary
Overall churn rate, model AUC, and optimal-threshold performance
The overall churn rate is 26.6%. The logistic regression + random forest pipeline achieves an AUC of 84.7%, indicating strong discrimination between churners and retained customers. At the Youden-optimal threshold of 0.3, model accuracy is 76.3%.
Churn Rate by Contract Type
Observed churn rate for each contract tier
Month-to-month contracts consistently show the highest churn risk among contract types. The top-churning contract tier is Month-to-month with an observed churn rate of 42.7%. Segments with fewer than 5 customers are excluded.
Churn Rate by Internet Service Type
Observed churn rate by internet service bundle
Fiber optic customers tend to churn at higher rates, possibly reflecting pricing dissatisfaction or heightened expectations. The service tier with the highest churn rate is Fiber optic at 42.1%.
Churn Rate by Payment Method
Observed churn rate by payment method — friction and engagement signal
Electronic check payers typically show the highest churn rate, while automatic payment methods (bank transfer, credit card) correlate with greater retention. The highest-churn payment method is Electronic check at 46%.
Logistic Regression Coefficients
Log-odds coefficients with significance flags — directional churn impact per predictor
Each bar shows the log-odds coefficient from logistic regression — positive values increase churn probability, negative values are protective. 11 predictor(s) are statistically significant at p < 0.05. The largest-magnitude predictor is Contract: Two Year.
Random Forest Variable Importance
MeanDecreaseGini importance from random forest — non-linear churn driver ranking
MeanDecreaseGini measures how much each variable reduces classification impurity across all trees in the random forest. Higher scores indicate more informative predictors, regardless of whether the effect is linear. The most important predictor by this measure is Total Charges.
Churn Rate by Tenure × Contract Type
Two-dimensional churn rate by tenure bucket and contract — reveals early-tenure risk
Each cell shows the observed churn rate for customers in a given tenure bucket (rows) and contract type (columns). Cells with fewer than 5 customers are excluded. Early-tenure month-to-month customers (0-12 months) typically show the highest churn rate — this is the highest-priority retention segment.
ROC Curve — Model Discrimination
Receiver Operating Characteristic curve showing AUC across all thresholds
The ROC curve plots the true positive rate (sensitivity) against the false positive rate (1 - specificity) at every classification threshold. The area under the curve (AUC) is 84.7%. A diagonal line represents random guessing (AUC = 50%). The Youden-optimal threshold is 0.3.
Confusion Matrix at Optimal Threshold
True vs predicted churn at the Youden-optimal classification threshold
At the Youden-optimal threshold of 0.3, the model achieves an overall accuracy of 76.3%. The matrix shows true churners (Churned/Churned), missed churners (Churned/Retained — false negatives), false alarms (Retained/Churned — false positives), and correctly retained customers (Retained/Retained). Minimising false negatives is typically the priority for churn prevention programs.