GenderPredictor 🧠

A simple neural network from scratch (in Python + NumPy) that predicts gender based on weight and height.
This project was built as a learning exercise to understand forward propagation, backpropagation, and gradient descent without using deep learning frameworks like TensorFlow or PyTorch.

---

📌 Features

• Implemented a 2 → 2 → 1 feedforward neural network from scratch.
• Uses the sigmoid activation function.
• Trains using Mean Squared Error (MSE) loss.
• Feature scaling for better convergence.
• Simple classification rule:
  • ≥ 0.5 → Female
  • < 0.5 → Male

---

📂 Project Structure

.
├── gender_predictor.py   # Main training & prediction script
├── README.md             # Project documentation
└── requirements.txt      # Dependencies list

---

🚀 Installation & Usage

1. Clone the repository

git clone https://github.com/yousifnimah/GenderPredictor.git
cd GenderPredictor

2. Install dependencies

pip install -r requirements.txt

3. Run training

jupyter notebook gender_predictor.ipynb

---

Example Output

Epoch  950 | loss: 0.0185
Epoch 1000 | loss: 0.0168

==== Gender Prediction Results ====
👤 Emily
   📏 Height : 160 cm
   ⚖️ Weight : 50 kg
   🎯 Prediction : 0.958 → Female

👤 Frank
   📏 Height : 190 cm
   ⚖️ Weight : 100 kg
   🎯 Prediction : 0.039 → Male
===================================

---

How It Works

1. Feature Scaling

   x_scaled = (x - mu) / sigma

   This centers the data around zero and reduces large value effects.

2. Feedforward

   • Inputs pass through hidden layer (2 neurons).
   • Output neuron gives probability via sigmoid.

3. Backpropagation

   • Manual gradient computation using chain rule.
   • Weight & bias updates via gradient descent.

4. Prediction

   if output >= 0.5:
       gender = "Female"
   else:
       gender = "Male"

---

Network Diagram

License

This project is licensed under the MIT License – you can freely use, modify, and distribute it.

---

🤝 Contributing

Pull requests are welcome!
Ideas for improvements:

• Add more features (e.g., age, BMI).
• Visualize the decision boundary.
• Save/load trained models.