ACCA7025 Application of Machine Learning Assignment Brief Jan 2025 |

TASK DESCRIPTION

Introduction

In this assignment, you will be using your knowledge of machine learning algorithms and methods to generate two ML models capable of classifying images. For this assignment, you are asked to write a report (3000 words), excluding references and appendices, in word or pdf format that includes an abstract, introduction, related work in the literature, methodology, results, discussion, conclusion, references and an appendix.

This report should be submitted alongside the scripts you developed. Additionally, you are asked to provide a short (approximately 5 minutes) demonstration video showing your code working and describing the various functions of the code. Instructions for uploading your report, code and video will be provided on Moodle and described in class.

You will submit:-

You will submit a written report as detailed above.
Include references using the School’s agreed standard (see referencing document on our Moodle course page).
All your source code as an appendix to your main report.
Your source code - uploaded to Moodle, separately.
A 5 minute video demonstrating your working code and highlighting key aspects of your code/model/results. Provide a link to this video in your report and email a link to your tutor. Alternatively upload a copy of your video with your source code, or email a copy to your tutor.

Task

You are expected review the literature and related work relevant to your task and then create at least two classifiers capable of classifying an image dataset. You must use Python-based scripts using a suitable ML package such as Scikit-Learn. Both of the ML classifiers should be selected from:

• Decision trees,
• SVMs,
• Naïve Bayes or
• KNNs 

Your aim is to keep the model as simple as possible without compromising the effectiveness of your classifier.

Important – avoid using convolutional neural networks in this assignment. 

You are expected to modify hyperparameters to get the best performance of each classifier, and to compare the results between classifiers. You will need to describe and justify the choice you make, and evaluate your ML models using established metrics such as F1-scores, accuracy, recall and precision as well as discussing confusion matrix results by, for example, explaining why a predicted class might have poor precision.

You have been introduced to several datasets that have been used in the literature to evaluate models. Initially, use the MNIST Fashion dataset provided for you in class to evaluate and compare your models, then use the more sophisticated Cifar10 image dataset, and re-train and re-evaluate your models.  Don’t forget to provide suitable graphs and tables that will clearly help illustrate your results, analyses and discussions.