This project simulates a real-world scenario where I act as a data analyst at Waze. The goal is to develop a machine learning model to predict user churn and generate actionable insights that support user retention strategies.
Problem Statement:
Waze wants to improve user retention by identifying users likely to stop using the app. Using behavioral engagement data, the task is to:
- Build a churn prediction model
- Evaluate different modeling techniques
- Communicate insights to technical and non-technical stakeholders
Key Insights:
- Driving days and days after onboarding are strong churn predictors.
- Churn behavior is not significantly different between iPhone and Android users.
- Early engagement is critical — churners often have fewer active days early on.
Business Recommendations:
- Focus on improving the first 30 days of user onboarding.
- Identify and re-engage low-driving-ratio users.
- Use a model to flag at-risk users monthly for proactive outreach.
Modeling and Evaluation:
The final model, XGBoost, outperformed others in identifying churners. Key metrics include:
- Accuracy: 74.3%
- F1 Score (Churn): 0.38
Link to GitHub Repository: Waze Customer Churn Prediction