Customer Purchase Prediction – ANN Model

This project builds and deploys an Artificial Neural Network (ANN) to predict whether a customer is likely to make a purchase, based on demographic and behavioral features.
The solution includes data preprocessing, model training, evaluation, and deployment with a simple app interface.

---

📂 Project Structure

.
├── customerData.csv                 # Original dataset (50,000 rows)
├── ann_customer_purchase_prediction.ipynb   # Main notebook (preprocessing + model training)
├── app.py                           # Streamlit app for deployment
├── train.ipynb                      # Training script (alternative to main notebook)
├── accuracy_loss.png                 # Training accuracy/loss curves
├── models/                          # (Optional) Saved models and encoders
│   ├── purchasePredictionModel.h5
│   ├── scaler.pkl
│   ├── label_encoder_gender.pkl
│   └── onehot_encoders_*.pkl

---

⚙️ Installation

1. Clone this repository:

   git clone https://github.com/your-username/customer-purchase-prediction.git
   cd customer-purchase-prediction

2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate   # Mac/Linux
   venv\Scripts\activate      # Windows

3. Install dependencies:

   pip install -r requirements.txt

---

📊 Dataset

• The dataset contains 50,000 entries of customer records.
• Features include demographics (Age, Gender, Region), behavior (Purchases, Time spent, Discounts availed), and loyalty information.
• Target: PurchaseStatus (0/1).

Synthetic expansion to 500k rows can also be generated for large-scale testing.

---

🧠 Model Architecture

• Input: Encoded + scaled feature set
• Hidden Layers:
  • Dense(128, ReLU, with L2/Dropout)
  • Dense(64, ReLU)
• Output: Dense(1, Sigmoid)
• Loss: binary_crossentropy
• Optimizer: Adam
• Callbacks: EarlyStopping (patience=10, restore best weights), TensorBoard

---

🚀 Training

Run the notebook or the training script:

jupyter notebook ann_customer_purchase_prediction.ipynb

or

python train.ipynb

The model and preprocessing objects are saved as:

• purchasePredictionModel.h5
• scaler.pkl
• label_encoder_gender.pkl
• onehot_encoders_*.pkl

---

🌐 Deployment (Streamlit App)

Run the app locally:

streamlit run app.py

This opens a web UI where you can enter customer details and get a prediction.

---

📈 Results

• Validation Accuracy: ~92.5%
• Validation AUC: ~0.93
• Stable training with minimal overfitting
• Example accuracy/loss curve:

---

🏷️ Use Cases

• Predict customer purchase likelihood
• Support loyalty program targeting
• Guide marketing campaign personalization
• Forecast sales in e-commerce/retail

---

🛠️ Requirements

• Python 3.9+
• TensorFlow 2.x
• scikit-learn
• pandas, numpy, matplotlib
• streamlit (for app)

Install with:

pip install tensorflow scikit-learn pandas numpy matplotlib streamlit

---

📌 Author

Developed by Gautham Vijayaraj.
Feel free to fork, improve, and use for your own ML projects!