🏦 Customer Churn Prediction – End-to-End ML Project

🚀 Live App:
👉 https://churnxgboost.streamlit.app/

---

📌 Project Overview

Customer churn is a critical business problem where companies lose customers to competitors.
This project builds an end-to-end Machine Learning pipeline to predict whether a customer will churn based on demographic and financial features.

The final deployed model uses XGBoost, selected after comparing multiple models.

---

🎯 Business Objective

• Identify customers likely to churn
• Enable targeted retention strategies
• Reduce customer acquisition cost
• Improve long-term revenue

---

📊 Dataset Information

• Total Records: 10,000
• Target Variable: Churn (0 = No, 1 = Yes)
• Class Distribution: ~80% Non-Churn, ~20% Churn
• Features:
  • Credit Score
  • Geography
  • Gender
  • Age
  • Tenure
  • Balance
  • Number of Products
  • Has Credit Card
  • Is Active Member
  • Estimated Salary

---

🔍 Exploratory Data Analysis (EDA)

• Identified class imbalance
• Analyzed feature distributions
• Detected and handled outliers using IQR
• Correlation analysis performed
• Key insights:
  • Germany showed higher churn rate
  • Inactive members churn more
  • Age positively correlates with churn

---

🧹 Data Preprocessing

• Removed duplicates
• OneHotEncoded categorical features:
  • Geography
  • Gender
• Applied StandardScaler (for distance-based models)
• Used Stratified Train-Test Split (80-20)
• Handled class imbalance using:
  • class_weight
  • scale_pos_weight in XGBoost

---

🤖 Models Implemented

The following models were trained and compared:

• Logistic Regression
• K-Nearest Neighbors (KNN)
• Support Vector Classifier (SVC)
• Decision Tree
• Random Forest
• XGBoost

Evaluation Metrics Used:

• Accuracy
• Precision
• Recall
• F1 Score
• Confusion Matrix

---

🏆 Final Model Selection

Selected Model: XGBoost

Reasons:

• Highest Recall (best churn detection)
• Strong F1 Score
• Balanced performance
• Controlled overfitting

---

🖥 Deployment

The final model was deployed using:

• Streamlit (Frontend + Backend)
• XGBoost model serialized using Pickle
• Hosted on Streamlit Cloud

🔗 Live App:
https://churnxgboost.streamlit.app/

---