AI/ML Engineering Internship - DevelopersHub Corporation

This project is a part of my AI/ML Engineering Internship at DevelopersHub Corporation, Islamabad.

Internship Details

• Company: DevelopersHub Corporation, Islamabad 🇵🇰
• Internship Period: July - September 2025

Customer Churn Prediction

Objective

This project implements an end-to-end machine learning pipeline for predicting customer churn. The goal is to identify customers who are likely to discontinue using a service, enabling proactive retention strategies.

• View the app screenshots here
• Link to App Online: click the link in the About section on the left

Features

• Data preprocessing and feature engineering
• Model training with hyperparameter tuning
• Web-based interface for predictions
• Input validation and error handling
• Model performance visualization

Installation

1. Clone the repository:

   git clone 'repository s URL'
   cd customer-churn-predictor

2. Create and activate a virtual environment:

   python -m venv churn-env
   source churn-env/bin/activate  # On Windows: churn-env\Scripts\activate

3. Install the required packages:

   pip install -r requirements.txt

Usage

1. Running the Web App:

   streamlit run app.py

   Open your browser and navigate to http://localhost:8501

Methodology / Approach

Data

• Dataset source - Kaggle

• Dataset: The model is trained on a dataset containing 50,000 customer records with the following features:

  • tenure_months: Number of months as a customer
  • monthly_usage_hours: Average monthly usage hours
  • has_multiple_devices: Whether the customer uses multiple devices (0/1)
  • customer_support_calls: Number of support calls made
  • payment_failures: Number of payment failures
  • is_premium_plan: Whether the customer is on a premium plan (0/1)
  • churn: Target variable (0 = No churn, 1 = Churn)

Preprocessing

• Handling missing values
• Feature scaling
• Encoding categorical variables
• Train-test split (80-20)

Model Training

• Implemented using scikit-learn's Pipeline API
• Models evaluated:
  • Logistic Regression
  • Random Forest Classifier
• Hyperparameter tuning using GridSearchCV
• Model evaluation using:
  • Accuracy
  • Precision
  • Recall
  • F1-Score
  • AUC-ROC

Web Interface

• Built with Streamlit
• Input validation
• Real-time predictions
• Model performance visualization

Key Results

Model Performance

Metric	Score
Accuracy	0.984
Precision	0.683
Recall	0.406
F1-Score	0.509
AUC-ROC	0.974

Key Observations

1. The Random Forest model outperformed Logistic Regression in terms of overall accuracy and AUC-ROC score.
2. The most important features for predicting churn were found to be:
   • tenure_months
   • monthly_usage_hours
   • payment_failures
3. The model shows good generalization with consistent performance on the test set.

Project Structure

ChurnPredictor/
├── app-screenshots/             # Snapshots of the App
├── dataset/                     # Dataset directory
│   └── customer_churn_dataset.csv
├── .gitignore                  # Git ignore file
├── README.md                   # This file
├── app.py                     # Streamlit web application
├── customer_churn_pipeline.ipynb # Jupyter notebook for model development
├── best_churn_model_pipeline.pkl # Trained model pipeline
├── model_info.pkl              # Model metadata and validation rules
└── requirements.txt            # Dependencies

Dependencies

• Python 3.8+
• pandas
• numpy
• scikit-learn
• streamlit
• matplotlib
• seaborn
• joblib