0.7946
Logistic Regression Accuracy
Analysis Outputs and Insights
Visual outputs generated from the churn pipeline, along with business interpretation of model results.
Model Snapshot
0.7584
Random Forest Accuracy
0.7887
Random Forest ROC-AUC
26.58%
Overall Churn Rate
19.80%
High-Risk Segment Share
Detailed Model Evaluation
- Logistic Regression Accuracy: 0.7946
- Random Forest Accuracy: 0.7584
- Random Forest Precision: 0.5491
- Random Forest Recall: 0.5080
- Random Forest F1 Score: 0.5278
- Random Forest ROC-AUC: 0.7887
Interpretation: Logistic Regression gives better overall accuracy, while Random Forest is used for richer non-linear signal discovery and feature importance analysis.
How To Read This Analysis
- The base dataset is Telco Customer Churn and is adapted to OTT/SaaS style business questions.
- Features such as usage frequency, last login days, and support calls are engineered proxies.
- Each visualization highlights one churn driver; model metrics validate predictive reliability.
- Risk segmentation converts model probabilities into actionable retention groups.
Visualization Outputs
Churn Distribution
Shows churn and non-churn customer counts with overall churn rate.
- What to observe: the gap between churned and retained user counts.
- Business meaning: around one in four users are leaving, which is materially high for subscription products.
- Action: make churn reduction a core KPI for product and customer success teams.
Tenure vs Churn
New users churn significantly more than loyal long-tenure users.
- What to observe: churn is highest in early lifecycle segments.
- Business meaning: onboarding quality and early value delivery are critical retention levers.
- Action: launch first-30-day and first-90-day intervention journeys.
Usage vs Churn
Lower usage intensity is associated with higher churn tendency.
- What to observe: churn decreases as engagement intensity increases.
- Business meaning: inactive users likely do not perceive enough product value.
- Action: trigger re-engagement nudges, tutorials, and habit-building campaigns.
Monthly Charges vs Churn
Higher monthly charges correlate with elevated churn risk.
- What to observe: churned users are concentrated at higher price ranges.
- Business meaning: perceived value may not justify cost for a subset of users.
- Action: test pricing bundles, loyalty discounts, and value-communication messaging.
Support Calls vs Churn
Frequent support interactions indicate higher dissatisfaction and churn risk.
- What to observe: churn rises with repeated support contacts.
- Business meaning: unresolved issues or product friction increase cancellation likelihood.
- Action: prioritize first-contact resolution and proactive outreach for repeated-ticket users.
Correlation Heatmap
Highlights positive and negative relationships between features and churn.
- What to observe: strength and direction of feature relationships.
- Business meaning: helps shortlist strongest churn-linked signals for intervention design.
- Action: use high-impact correlated features to design rule-based churn alerts.
Model Accuracy Comparison
Compares predictive performance between Logistic Regression and Random Forest.
- What to observe: benchmark difference between linear and ensemble models.
- Business meaning: model selection can balance interpretability and flexibility.
- Action: use Logistic Regression for explainability and Random Forest for signal exploration.
Feature Importance
Top drivers include monthly charges, tenure, and support-related behavior.
- What to observe: rank order of features influencing predicted churn.
- Business meaning: highlights where product, pricing, and CX teams should focus first.
- Action: align retention campaigns to the top five drivers for measurable impact.
Top Churn Signals Identified
Monthly Charges
Most influential feature. Higher monthly billing aligns with higher churn probability.
Tenure
Low-tenure customers are more vulnerable to churn, especially in early lifecycle stages.
Support Behavior
Rising support interaction levels are a strong dissatisfaction and churn warning signal.
Risk Segmentation Logic
Low Risk
Churn probability <= 0.40. Maintain with regular communication and loyalty offers.
Medium Risk
Churn probability between 0.40 and 0.70. Engage with targeted usage nudges.
High Risk
Churn probability > 0.70. Prioritize proactive retention, incentives, and human support.
Key Business Recommendations
- Offer targeted discounts to high-risk monthly-plan customers.
- Improve onboarding quality in the first 12 months.
- Launch engagement campaigns for low-usage users.
- Provide proactive support for users with repeated service issues.
Execution priority: Start with High-Risk monthly-plan customers, then roll out Medium-Risk engagement workflows.
Assumptions and Limitations
- Some OTT-style behavior fields are inferred from telecom service attributes, not direct app telemetry.
- Results are valid for this dataset distribution and should be re-evaluated for other markets or periods.
- Threshold-based segmentation can be tuned based on business cost of churn and campaign budget.