📡
ChurnScope
IBM Watson Telco — Churn Prediction System
ML MODEL ACTIVE
🎯 Predict Churn
📊 Model Comparison
💡 Model Insights
Customer Churn Prediction
① Charges & Tenure
Tenure (months)
Monthly Charges ($)
② Contract Type
Contract
Month-to-month
One year
Two year
③ Internet & Protection Services
Internet Service
Fiber optic
DSL
No
Tech Support
No
Yes
Online Security
No
Yes
Online Backup
No
Yes
Device Protection
No
Yes
④ Billing Setup
Payment Method
Electronic check
Mailed check
Bank transfer (automatic)
Credit card (automatic)
Paperless Billing
Yes
No
⑤ Customer Profile
Senior Citizen
No
Yes
Partner
No
Yes
Dependents
No
Yes
⚡ Run Prediction
🤖
Fill in customer details and click
Run Prediction
Model Prediction
✅ NOT CHURN
0%
CHURN PROBABILITY
LOW RISK
Customer likely to stay
Risk Factors
Model Comparison
🔵
Logistic Regression
Accuracy
73.2%
Precision (Churn)
50%
Recall (Churn)
79%
F1 (Churn)
61%
ROC-AUC
83.7%
CV ROC-AUC
84.9%
🟢
Random Forest
Accuracy
75.6%
Precision (Churn)
53%
Recall (Churn)
76%
F1 (Churn)
63%
ROC-AUC
83.7%
CV ROC-AUC
84.9%
🏆 Best Model
🟠
Gradient Boosting
Accuracy
74.2%
Precision (Churn)
51%
Recall (Churn)
78%
F1 (Churn)
62%
ROC-AUC
84.0%
CV ROC-AUC
84.6%
🟡
XGBoost
Accuracy
74.1%
Precision (Churn)
51%
Recall (Churn)
78%
F1 (Churn)
62%
ROC-AUC
83.9%
CV ROC-AUC
84.5%
ℹ️ Best model ranked by ROC-AUC (test set). CV ROC-AUC from 5-fold StratifiedKFold. Gradient Boosting trained with
sample_weight='balanced'
.
Metric Comparison Across All 4 Models
Model Insights
Top Features Driving Churn — Gradient Boosting 🏆
= engineered feature
⚠️ High-Risk Customer Signals
61%
FiberNoAddons
Fiber optic, zero add-ons. Highest churn signal in dataset.
54%
NoProtection
No security, backup, or tech support. Easy to walk away.
53%
HighRiskCombo
Month-to-month + above median charges.
50%
EasyLeaver
Paperless billing + electronic check. Most frictionless exit.
53%
IsNewCustomer
First 6 months — critical retention window.
34%
SoloCustomer
No partner, no dependents. More likely to shop around.
Confusion Matrix — Gradient Boosting 🏆
Test set · Best model by ROC-AUC (0.840)
934
True Negative
Correctly stayed
100
False Positive
Predicted churn, stayed
131
False Negative
Missed churner ⚠️
244
True Positive
Caught churner ✅
Predicted Stay
Predicted Churn