Machine Learning Coursework 2 – Neural Networks & Model Evaluation

📌 Overview

This project was developed for COMP6011: Machine Learning (Coursework 2).

The aim was to design, implement, and evaluate perceptron-based classifiers and compare them with other machine learning algorithms (kNN, Naïve Bayes) using the Palmer Penguins dataset.

Key goals included:

• Implementing a Multi-Class Perceptron and a Multi-Layer Perceptron (MLP).
• Exploring hyperparameter tuning (epochs, learning rate, train/test split).
• Comparing performance with k-Nearest Neighbours (kNN) and Naïve Bayes.
• Critically evaluating model efficiency, effectiveness, and limitations.

▶️ Run the Notebook

You can run the project interactively on Google Collab:

Google Colab

🛠 Implementation

Multi-Class Perceptron

• Best epoch: 600 → ~80% accuracy.
• Best learning rate: 0.164 → ~66% accuracy.
• Best split ratio: 90% train / 10% test → ~80.6% accuracy.

Multi-Layer Perceptron (MLP)