ANN-Classification-churn

This project builds an Artificial Neural Network (ANN) to predict customer churn using the "Churn_Modelling.csv" dataset. It includes complete preprocessing, model training, and a simple web app interface to make real-time predictions.

---

📌 Project Overview

Customer churn prediction is vital for improving customer retention. Using a multi-layer ANN, this project learns patterns from customer data (like credit score, geography, age, balance, etc.) to classify whether a customer is likely to churn or stay.

---

📂 Project Structure

.
├── Churn_Modelling.csv           # Dataset
├── app.py                        # Web app for real-time predictions
├── experiments.ipynb             # Data exploration, preprocessing, model training
├── prediction.ipynb              # Manual prediction walkthrough using saved artifacts
├── model.h5                      # Trained ANN model
├── scaler.pkl                    # StandardScaler for numeric features
├── label_encoder_gender.pkl      # LabelEncoder for 'Gender'
├── onehot_encoder_geo.pkl        # OneHotEncoder for 'Geography'
├── requirements.txt              # Python dependencies
└── README.md                     # Project documentation

---

🔍 Dataset Description

• Source: [Churn_Modelling.csv](https://www.kaggle.com/datasets/shubhendra7/customer-churn-prediction)
• Size: 10,000 rows × 14 columns
• Target variable: Exited (1 → churned, 0 → retained)

Key features used:

• CreditScore
• Geography (France, Spain, Germany)
• Gender
• Age
• Tenure
• Balance
• NumOfProducts
• HasCrCard
• IsActiveMember
• EstimatedSalary

---

🛠️ Tech Stack

• Language: Python 3.x

• Libraries:

  • TensorFlow / Keras – Model development
  • pandas, numpy – Data processing
  • scikit-learn – Preprocessing, encoders
  • joblib – Model and encoder saving
  • Streamlit – Web interface

---

🚀 How to Run

1. Clone the Repository

git clone https://github.com/Github-Shashwat/ANN-Classification-churn.git
cd ANN-Classification-churn

2. Install Dependencies

It’s recommended to use a virtual environment (e.g., venv or conda).

pip install -r requirements.txt

3. Train the Model (Optional)

If you want to retrain the model:

• Open experiments.ipynb
• Follow the steps: EDA → preprocessing → training
• The trained model and encoders will be saved as .h5 and .pkl files

4. Run the Web App

streamlit run app.py

Visit http://localhost:8501 in your browser to access the interface.

---

📈 Model Architecture

• Input layer: 11 features

• Hidden layers:

  • Dense(6) with ReLU
  • Dense(6) with ReLU

• Output layer:

  • Dense(1) with Sigmoid

Loss: binary_crossentropy Optimizer: adam Metric: accuracy

---

🎯 Prediction Pipeline

1. User enters inputs via the web form.
2. Categorical features (Gender, Geography) are encoded using saved .pkl files.
3. Numerical features are scaled using the stored StandardScaler.
4. The ANN model makes a prediction: churn probability and class label (0 or 1).

---

🔍 Example Output

Input:
- Geography: France
- Gender: Female
- Age: 40
- Balance: 50000
...

Output:
- Churn Probability: 0.76
- Prediction: High Risk of Churn (1)

---

📊 Evaluation Metrics

Check experiments.ipynb for:

• Accuracy
• Confusion Matrix
• Precision, Recall, F1-score
• Training/Validation Loss curves

---

✅ Future Improvements

• Add SHAP/Explainable AI visualizations
• Deploy app using Docker or Cloud (AWS/GCP)
• Enable CSV upload for batch prediction
• Add cross-validation and hyperparameter tuning