Project Overview

Customer churn remains one of the most significant operational and financial challenges facing subscription-based businesses. Retaining an existing customer is considerably less expensive than acquiring a new one, making churn prediction an important business intelligence capability.

The Customer Churn Intelligence Platform was developed as a complete Machine Learning solution capable of predicting customer churn using historical behavioural data while simultaneously providing decision-makers with interpretable insights and retention recommendations.

• End-to-end supervised Machine Learning workflow
• Comprehensive exploratory data analysis
• Automated data preprocessing using Scikit-Learn Pipelines
• Multiple model comparison and evaluation
• Business-focused interpretation of predictions
• Interactive Streamlit deployment

Business Problem

Customer attrition directly reduces recurring revenue while increasing marketing and acquisition costs. Organizations frequently struggle to identify customers who are most likely to discontinue their services before it becomes too late.

This project addresses that challenge by developing an intelligent classification model capable of identifying high-risk customers using demographic information, subscription history, contract characteristics and service usage patterns.

Dataset Overview

The project utilizes the IBM Telco Customer Churn dataset, one of the most widely used benchmark datasets for binary classification problems involving customer retention.

• 7043 customer records
• 19 predictive variables
• Binary classification target (Churn)
• Customer demographics
• Contract information
• Internet subscription details
• Billing information
• Customer support services

Machine Learning Workflow

The project follows a structured end-to-end Machine Learning workflow, beginning with raw customer data and ending with an interactive decision-support application deployed using Streamlit.

1. Business Problem Definition
2. Exploratory Data Analysis (EDA)
3. Data Cleaning & Preprocessing
4. Feature Engineering
5. Model Development
6. Hyperparameter Optimization
7. Model Evaluation
8. Business Interpretation
9. Interactive Streamlit Deployment

Exploratory Data Analysis

Before model development, extensive exploratory analysis was performed to understand customer behaviour, identify relationships among variables, detect anomalies, and uncover the major factors associated with churn.

• Approximately 26.6% of customers had churned, indicating a moderately imbalanced classification problem.
• Customers with month-to-month contracts demonstrated substantially higher churn rates than those with one-year or two-year agreements.
• Customers with shorter tenure were considerably more likely to discontinue their subscriptions.
• Higher monthly service charges showed a positive association with churn behaviour.
• Lack of technical support services was associated with an increased probability of customer attrition.

Model Development & Evaluation

Multiple supervised Machine Learning algorithms were trained, evaluated and compared using identical preprocessing pipelines to identify the most reliable model for customer churn prediction.

All categorical variables were encoded using One-Hot Encoding, numerical variables were standardized, and preprocessing was fully automated using a Scikit-Learn Pipeline.

Although more complex ensemble algorithms were evaluated, Logistic Regression achieved the strongest balance between predictive performance, computational efficiency, and interpretability.

Feature Importance & Business Insights

One of the major strengths of Logistic Regression is its interpretability. Model coefficients were analyzed to identify the strongest drivers of customer churn and translate statistical findings into actionable business recommendations.

Key Findings

• Customer Tenure was the strongest predictor of churn. Customers with shorter relationships were substantially more likely to leave.
• Customers on Month-to-Month Contracts exhibited the highest churn risk.
• Two-Year Contracts significantly improved customer retention.
• Customers paying higher Monthly Charges demonstrated increased churn probability.
• Lack of Technical Support was associated with elevated customer attrition.

Business Recommendations

• Introduce incentives that encourage migration from month-to-month subscriptions to long-term contracts.
• Implement proactive engagement campaigns during the customer's first year of service.
• Closely monitor customers with high monthly bills and provide loyalty offers before contract renewal.
• Improve awareness and adoption of technical support services to increase customer satisfaction.

Interactive Streamlit Dashboard

To transform the predictive model into a practical business application, the project was deployed using Streamlit as an interactive Customer Churn Intelligence Platform.

Application Features

• Interactive executive dashboard
• Single customer churn prediction
• Batch prediction using CSV upload
• Prediction probability and risk classification
• Downloadable prediction results
• Business-focused retention recommendations
• Interactive visual analytics

Technology Stack

Future Improvements

• SHAP-based explainable AI visualizations.
• Real-time REST API deployment.
• CRM integration for automated retention campaigns.
• Email notification for high-risk customers.
• Cloud deployment using Docker and Azure/AWS.
• Advanced ensemble learning and AutoML experimentation.