Customer Churn Prediction & Analysis

Project Overview

This project focuses on analyzing and predicting telecom customer churn using machine learning (Python) and business intelligence (Power BI).
The goal is to identify customers at risk of leaving, understand churn drivers, and provide actionable insights to reduce churn.

• Dataset: Adapted from Indian telecom data, customized to reflect U.S. states and demographics.
• Tools: Python (scikit-learn, pandas, matplotlib), Power BI
• Deliverables:
  • ML Model (Churn_Prediction.ipynb)
  • Power BI Dashboard (Churn Analysis.pbix)
  • Business Insights (README + narrative)

---

Machine Learning Model

The predictive model was built with a Random Forest Classifier.

Results:

• Accuracy: 85%
• Precision (Churners): 83%
• Recall (Churners): 64%
• F1-score (Churners): 0.73

These results show the model is strong at correctly predicting churners (precision), but there’s room to improve recall (capturing more actual churners).

Confusion Matrix & Classification Report available in the notebook.

---

Power BI Dashboard

The Power BI dashboard complements the ML model by providing interactive analysis of churn by demographics, contracts, services, and geography.

It includes:

• Churn by Gender, Age, State, Payment Method, and Contract Type
• Tenure group comparisons
• Churn by services (e.g., Internet type, add-ons)
• Predicted churner profiles with customer IDs

Dashboard Screenshots

Summary Dashboard

Prediction Dashboard

---

Insights & Narrative

Key findings from the dashboard and Power BI AI insights:

• Age & Tenure:
  Customers over 50 had the highest churn volume (2,838), accounting for 44.22% of total churners.
  Churn diverged most among customers with tenure ≥ 24 months, where churn volume (2,087) was significantly higher.

• Gender:
  Female churners (1,111) were higher than male churners (621).

• Contracts & Payment:
  Month-to-month contracts showed the highest churn rate (46.5%), while mailed check payments also had higher churn compared to other payment methods.

• Internet Services:
  Fiber optic users had the highest churn rate (41.1%), followed by cable (25.7%) and DSL (19.4%).

• Geography:
  Pennsylvania had the highest churn rate (57.2%), with Texas and Maryland following.

What I Learned

• How to preprocess categorical data and encode features for ML models.
• How to handle imbalanced churn data and interpret precision/recall trade-offs.
• Practical skills in Power BI: creating slicers, dynamic visuals, and integrating narratives using AI insights.
• How to combine machine learning + BI dashboards for storytelling and business impact.