Neural Networks

Overview

This repository contains the code and documentation: Neural Networks, which explores the implementation of an artificial neural network (ANN) for classification tasks. The project covers manual perceptron implementation, gradient descent optimization, and Scikit-learn-based neural networks.

Features

• Perceptron Implementation: Manually coded perceptron with a sigmoid activation function.
• Gradient Descent Optimization: Custom loss function and weight updates.
• Scikit-learn MLP: Utilizes a multi-layer perceptron (MLP) for classification.
• Model Fitting Challenge: Feature engineering applied to a challenging dataset.

Repository Structure

Neural-Networks - README.md # Project Documentation (this file) - Etivity2_NeuralNetworks_23233729.ipynb # Jupyter Notebook with full implementation - t4_data.csv # Dataset for model fitting challenge - Figures (Diagrams & Plots) - perceptron.png # Perceptron structure visualization - perceptron_xy.png # Perceptron dataset visualization - ann_2layers.png # ANN model architecture

Installation & Setup

1. Clone the Repository git clone https://github.com/yourusername/Neural-Networks.git cd Neural-Networks
2. Install Dependencies pip install numpy matplotlib scikit-learn
3. Run the Jupyter Notebook Launch Jupyter Notebook and open the provided .ipynb file: jupyter notebook Etivity2_NeuralNetworks_23233729.ipynb

Implementation Details

Task 1: Manual Perceptron Implementation

• Created a single-node Perceptron with a sigmoid activation function.
• Initialized the Perceptron with random weights and bias.
• Developed a group prediction function.
• Evaluated accuracy and visualized the group classification. Task 2: Gradient Descent Optimization
• Implemented a loss function (sum of squared errors).
• Created a gradient descent function for weight updates.
• Fitted the model until achieving a satisfactory performance.
• Evaluated final model accuracy and plotted predictions.

Task 3: Scikit-learn MLP Implementation

• Split dataset into training and testing sets.
• Defined and trained a multi-layer perceptron (MLP) model.
• Printed the mean accuracy of the trained model.
• Plotted model predictions for visual evaluation.

Task 4: Model Fitting Challenge

• Applied feature engineering techniques to improve model performance.
• Optimized the ANN structure for handling complex datasets.

Example Output

  Model: Scikit-learn MLP
  Training Accuracy: 95%
  Testing Accuracy: 91%
  Loss Reduction: Achieved optimal convergence

Key Learnings

• Developed a strong foundation in neural network architectures.
• Implemented a perceptron from scratch using NumPy.
• Improved model performance using gradient descent optimization.
• Learned to leverage Scikit-learn’s MLP for rapid ANN deployment.