Customer Churn Analysis

Problem Statement

Customer retention is a critical challenge for telecom service providers. Losing customers (churn) leads to significant revenue loss and increased acquisition costs. The goal is to identify customers at risk of leaving by analyzing historical customer behavior and service data, enabling timely intervention to improve retention.

Project Goal

To develop a predictive system using machine learning that classifies customers as likely to churn or not based on attributes like tenure, monthly charges, age, and gender. The final model is integrated into a user-friendly web application where predictions can be made instantly.

Key Features

• 🔍 Data Analysis & Visualization:

  • Conducted EDA to uncover patterns in churn behavior.

  • Found that churned customers had shorter tenure and higher monthly charges on average.

  • Explored churn by gender, contract type, and internet service.

• 🧪 Feature Engineering:

  • Converted categorical variables like gender and churn into binary numeric values.

  • Standardized continuous features using StandardScaler.

• 🤖 ML Model Development:

  • Built and compared several ML models:

    • Logistic Regression

    • K-Nearest Neighbors

    • Support Vector Machine (SVM) → Best performing

    • Decision Tree

    • Random Forest

  • Best model (SVM) achieved an accuracy of ~89% on the test set.

• 💻 Interactive Web App (Streamlit):

  • User inputs: Age, Gender, Tenure, and Monthly Charges.

  • Output: Real-time prediction (Yes or No) on whether the customer will churn.

Outcome

• Built a high-accuracy (~89%) churn prediction model.

• Integrated the model into a fully functional Streamlit app for instant predictions.

• Demonstrated an end-to-end ML workflow: EDA → Modeling → Deployment.

• Project showcases practical application of classification models to real-world business problems in telecom.