📊 Customer Churn Prediction — Telco Dataset

An end-to-end Machine Learning solution acieving 82% accuracy to predict customer churn for a telecom company.
This project covers data cleaning, feature engineering, model development, performance evaluation, and a deployment-ready ML pipeline.

---

🔍 Project Overview

• Objective: Predict whether a customer will churn (0/1) based on demographics, billing patterns, and service usage.
• Dataset: Telco Customer Churn dataset.
• Outcome:

✔️Achieved strong churn prediction accuracy using XGBoost (~82%).

✔️Identified key churn drivers such as contract type, monthly charges, tenure, and number of services.

✔️Built a fully automated ML pipeline (preprocessing + model) ready for real-world deployment.

✔️Reduced model preprocessing errors by implementing One-Hot Encoding and scaling inside a unified pipeline.

✔️Deployed a functional Flask web app allowing users to input customer details and get real-time churn predictions.

✔️Improved data quality by handling missing values and converting inconsistent fields like TotalCharges.

✔️Enabled business decisions by highlighting high-risk customers and improving retention strategies.

---

📊 Dashboard Insights (Power BI)

✔️Total customers analyzed: 7,030 with an average monthly cost of $64.80.

✔️73.42% customers stayed, while 26.58% churned — showing moderate churn risk.

✔️Fiber Optic users show the highest churn, indicating dissatisfaction with internet quality or cost.

✔️Customers without Tech Support churn significantly more, proving support availability impacts retention.

✔️Month-to-month contract customers churn the most, confirming contract length strongly influences loyalty.

✔️Security features (OnlineSecurity & OnlineBackup) show a clear trend: customers lacking these services churn at a much higher rate.

✔️Churn is comparatively higher across both genders, roughly similar between male and female customers.

---

🧠 Key Steps in the Project

✔️ 1. Data Cleaning & Preparation

• Handled missing or incorrect values (e.g., TotalCharges).
• Converted data types appropriately.
• Selected important features:
  • gender, SeniorCitizen, Partner, Dependents, tenure
  • Contract, PaymentMethod, InternetService
  • AvgMonthlyCost, NumServices, PaperlessBilling

✔️ 2. Feature Engineering

• Created domain-relevant features (AvgMonthlyCost, NumServices).
• Encoded categorical variables using One-Hot Encoding.
• Scaled numerical features using StandardScaler.

✔️ 3. Model Training

• Tested multiple models: Logistic Regression,KNN,Decision Trees, Random Forest, XGBoost.
• Selected XGBoost as the final model due to best accuracy and balance between precision/recall.
• Used a full Scikit-Learn Pipeline combining preprocessing + model.

✔️ 4. Evaluation

• Evaluated using Accuracy, Precision, Recall, F1-Score, Confusion Matrix, ROC-AUC.
• Final model shows strong churn prediction performance.

✔️ 5. Deployment-Ready Pipeline

• Saved model as:
  xgb_churn_pipeline.joblib
• Fully compatible with Flask API for production deployment.

---

📦 Project Structure

Customer-Churn-Analysis/

                   ├── Analysis using PowerBI/PowerBi file

                   ├── outputs/

                              └── output1.png

                              └── output2.png


                   ├── app/

                          └── app.py

                          └── templates/

                                       └── index.html

                          └── static/

                                    └── style.css

                          └── requirements.txt

                   ├──Customer Churn Analysis.ipynb

                   ├── README.md

                   └── telco_customer_churn.csv

---

⚙️ How to Run Locally

1. Clone the repository

git clone https://github.com/UK183/Customer-Churn-Analysis.git
cd Customer-Churn-Analysis

2. Create a virtual environment

python -m venv venv

3. Activate the environment

Windows:

venv\Scripts\activate

Mac/Linux:

source venv/bin/activate

4. Install Dependencies

pip install -r requirements.txt

5. Run the flask app

cd app
python app.py

Open in browser:

http://127.0.0.1:5000/

📈 Model Performance Model Result XGBoost (Final Model) ≈ High accuracy (replace with your score)

See the notebook for full evaluation metrics: Precision, Recall, F1-score, Confusion Matrix, ROC-AUC.

🧰 Tools & Technologies

• Python, Pandas, NumPy

• Scikit-Learn (Pipeline, Preprocessing)

• XGBoost

• Matplotlib, Seaborn

• Flask (for deployment)

• Jupyter Notebook

💼 Business Value

This project helps telecom companies:

• Identify customers likely to churn

• Reduce revenue loss

• Improve customer retention strategies

• Target high-risk customers with personalized offers

• Build automated churn prediction systems and dashboards

🚀 Future Enhancements

• Add SMS/Call usage patterns

• Implement advanced hyperparameter tuning

• Deploy using Docker / AWS / Azure

---

👤 Author

Kazi Umar
Linkedin profile: https://www.linkedin.com/in/umar-kazi18
💼 Data Analyst | ML Engineer | Data Science & AI Enthusiast | Power BI | Python | SQL