L-Layer Neural Network from Scratch (NumPy)

This project is a complete implementation of an L-layer neural network from scratch using only NumPy.
The goal was to learn and demonstrate the fundamentals of deep learning by coding everything manually, without using libraries like TensorFlow or PyTorch.

What I Built

From scratch, I implemented all the key components of a deep neural network:

• Parameter Initialization
• Forward Propagation
  • Linear step
  • Activation functions: ReLU (hidden layers), Sigmoid (output layer)
• Cost Function (Binary Cross-Entropy)
• Backward Propagation
  • Gradients for each layer
• Parameter Updates (Gradient Descent)
• Prediction Function
• Accuracy Calculation
• Training Loop (L_layer_model function with cost tracking)

Architecture

The network structure is flexible: you can define any number of layers and neurons with a Python list.

Example: layers_dims = [2, 10, 5, 1]

Input layer → 2 features

Hidden layer 1 → 10 neurons (ReLU)

Hidden layer 2 → 5 neurons (ReLU)

Output layer → 1 neuron (Sigmoid for binary classification)