Predictive Infrastructure

SentinelLTV.

A production-grade churn prediction engine utilizing XGBoost with GridSearch optimization to forecast customer retention with 80% accuracy.

Accuracy 80.0%
Mean ROC AUC 0.85
Inference <20ms
Optimized GridSearch
Feature Importance

01. Behavioral Drivers

Analysis reveals that Month-to-Month contracts and Fiber Optic service are the primary predictors of churn. This allows for targeted retention marketing based on contract structure.

Hyperparameter Tuning

Model performance was maximized via extensive GridSearch, identifying the optimal learning rate and tree depth to prevent overfitting.

Best Params: learning_rate: 0.01, max_depth: 3, n_estimators: 300
Validation Curve
Churn Distribution

Dataset Balance (26.5% Churn)

CV Distribution

Cross-Validation Stability

Confusion Matrix

Final Test Evaluation

XGBoost Classifier GridSearchCV Stratified K-Fold Imbalance Handling