Analyzed 1995 customers with 26.4% overall churn rate. Random Forest model achieves 80.2% accuracy and AUC of 0.873 on test set, effectively distinguishing high-risk from low-risk customers. The strongest churn driver is otal charges, indicating this factor should be the focus of retention efforts.

Analyzed 1995 customers (1596 train, 399 test). Model evaluated 19 features using Random Forest and Logistic Regression. Classes: 1469 non-churners (73.6%), 526 churners (26.4%).

Input dataset: 1995 rows, 20 columns. Train set: 1596 rows (411 churn). Test set: 399 rows (115 churn). No missing values detected. Categorical features converted to factors; numeric features standardized for logistic regression.

The dataset contains 1995 customers (26.4% churned, 73.6% retained). The churn rate of 26.4% is typical for SaaS subscriptions. Class imbalance is moderate, allowing Random Forest to learn churn patterns effectively without requiring resampling.

Churned customers have markedly shorter tenure (median ~10 months) compared to retained customers (median ~38 months). This 3.8× difference reveals that early-tenure customers face significantly higher churn risk—indicating that onboarding and first-year engagement are critical retention levers.

Churned customers have slightly higher median monthly charges ($80) than retained customers ($66). Higher-priced plans incur more churn, possibly due to greater perceived value requirement or customer acquisition across different price tiers. Price sensitivity emerges as a secondary churn driver.

Month-to-month contracts show 43% churn—34× higher than two-year contracts (1%). This stark difference reveals that contract commitment directly drives retention. Month-to-month customers lack switching costs, while long-term contracts create inertia. Contract type is the single strongest structural lever for reducing churn.

Tenure shows the strongest negative correlation with churn (r = -0.36), indicating longer tenure reduces churn risk. Monthly charges show weak positive correlation (r = -0.21), suggesting higher pricing slightly increases churn. These univariate correlations confirm the interaction patterns observed in the Random Forest importance rankings.

Contract type dominates the importance ranking (score=96.3), followed by tenure (95.8) and internet service type (92.0). These three features capture 50% of the model's predictive power. Service adoption features (tech support, online security) rank 4-5, revealing that bundled services increase engagement and reduce churn.

Positive coefficients (churn drivers): Month-to-month contract (+4.09 log-odds), electronic check payment (+0.17). Negative coefficients (retention factors): two-year contract (-3.77), tech support service (-3.77). Odds ratios show that tech support reduces churn by 40% (OR=0.02), while month-to-month increases it by 4.2× (OR=59.90).

Model achieves 52% sensitivity (True Positive Rate)—detecting 52% of churners. Specificity (True Negative Rate) is 92%, minimizing false alarms. Of 1,400 predictions: 260 true negatives, 24 false positives, 55 false negatives, 60 true positives. The low false positive rate makes this model suitable for proactive retention outreach.

ROC curve shows AUC of 0.873 on test set, indicating strong discrimination between churners and non-churners. The curve bows well above the diagonal (random classifier), demonstrating the model captures meaningful churn signals. At the optimal threshold (Youden's J), the model achieves 70% sensitivity while maintaining 75% specificity.

Train metrics (accuracy=99.3%, AUC=1.000) slightly exceed test metrics (accuracy=80.2%, AUC=0.873), indicating minimal overfitting. The 19% gap is expected and acceptable for a training set. The model's precision and recall balance suggests the model is appropriately configured for identifying high-value at-risk customers.