Arekflo2002/Project_Computational_Intelligence_Part2

🧠 Neural Network for Diabetes Prediction

This project implements a custom Artificial Neural Network (ANN) from scratch in Python to predict diabetes based on medical data.

📌 Features

✔️ Custom implementation of a fully connected feedforward ANN
✔️ Backpropagation algorithm for training
✔️ Configurable hyperparameters (epochs, learning rate, activation functions, layers, etc.)
✔️ Training and evaluation using the diabetes.csv dataset
✔️ Visualization of learning curves and model predictions

🏗 Implementation Details

The neural network is implemented using NumPy, without external deep learning frameworks like TensorFlow or PyTorch.

🔹 Model Architecture

• Input Layer: 8 features from the dataset
• Hidden Layers: Configurable number of layers and neurons
• Activation Functions: Sigmoid, ReLU, Tanh (configurable)
• Output Layer: Single neuron with Sigmoid activation (binary classification)

🔹 Training Process

1. Forward Propagation: Compute activations for each layer.
2. Loss Calculation: Binary cross-entropy loss is used.
3. Backward Propagation: Gradients are computed and weights are updated.
4. Parameter Updates: Using gradient descent with a configurable learning rate.

🔹 Configurable Hyperparameters

Modify parameters in config.py or directly in main.py:

EPOCHS = 10000
LEARNING_RATE = 0.01
LAYERS = [8, 3, 3, 1]  # Input, Hidden, Output
ACTIVATION_FUNCTION = 'sigmoid'  # Options: 'sigmoid', 'relu', 'tanh'

📖 References

• Research Paper 1
• Research Paper 2
• Research Paper 3