Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit

COMP3003

MACHINE LEARNING

Overview

This document contains all the necessary information pertaining to the assessment of COMP3003 Machine Learning. The module is assessed via 100% coursework, across two elements: 30% Set Exercises and 70% Report.

The sections that follow will detail the assessment tasks that are to be undertaken. The submission and expected feedback dates are presented in Table 1. All assessments are to be submitted electronically via the respective DLE module pages before the stated deadlines.

Table 1: Assessment Deadlines

All assessments will be introduced in class to provide further clarity over what is expected and how you can access support and formative feedback prior to submission. Whilst the assessment information is provided at the start of the module, it is not necessarily expected you will start this immediately – as you will often not have sufficient understanding of the topic. The module leader will provide guidance in this respect.

Assessment 1: Set Exercises

This assignment contributes to 30% of the overall module mark for COMP3003 and is an individual assignment.

The focus of this coursework is to assess your understanding of unsupervised machine learning techniques. You are required to write matlab code to implement the Kmeans clustering algorithm from 1st principles. This is an extension of Lab 4 on Kmeans clustering. The detailed tasks are as below.

Write matlab code to implement the Kmeans clustering algorithm from 1st principles. You should demonstrate clustering for 3 clusters with each cluster created based on a 2D uncorrelated Gaussian distributed dataset. You are free to make assumptions on the number of points in each cluster and the degree of overlapping among clusters. The assumptions should be clear enough to demonstrate the concept of Kmeans clustering and clearly show the limitations of Kmeans approach (through your explanation).

You are required to write a report to cover the tasks above. The individual report should be no more than 1,500 words (excluding diagrams, images, tables, matlab code/comments, and references). The report should be organised as follows:

- Abstract: about 150 words

- Introduction: unsupervised learning and Kmeans clustering – about 450 words

- Implementation: implementation of Kmeans clustering (including implementation steps, matlab code, results/screenshots, performance of clustering, and explanations) – about 600 words

- Discussions and conclusions (including drawbacks of Kmeans clustering) – about 300 words

- References

- Appendix (including all your Matlab code)

Any references should be appropriately cited in the report. Harvard referencing style is recommended. You should write your report as concisely as possible and it is important that you do not exceed or within 10% of the allowed word limit.

Assessment Criteria:

The report will be assessed based on the following criteria.

- Abstract, Presentation, Structure, Conclusions, References (20%): are the abstract and conclusions described appropriately? Are discussions sufficient? Is the report well-presented and structured? Are the references appropriate?

- Introduction of unspervised machine learning and K-means (20%): are unsupervised learning and K- means clustering concepts explained appropriately? Is it easy to follow? Is it clear?

- Implementation (60%): Are the implementation steps clearly described? Is evidence (e.g. screenshots) sufficient? Have the results been explained and/or analysed appropriately? Is the matlab code well explained and appropriate?

Assessment 2: Project Report

This assignment contributes to 70% of the overall module mark for COMP3003 and is an individual assignment.

The focus of this coursework is to assess your understanding of machine learning techniques, with a focus on supervised learning and reinforcement learning, and their practical applications. You are required to (1) undertake a literature review on the above two machine learning techniques and identify two real-world applications for each approach; (2) write matlab code to develop supervised learning models (neural networks) for intrusion detection in network security applications. The detailed tasks are as below.

Task 1: Literature review

Undertake a review of machine learning techniques, with a focus on supervised learning techniques. Understand differences between supervised learning and reinforcement learning. Identify two real- world applications for each approach.

Task 2: Write Matlab code to build a network intrusion detector based on supervised learning. You will develop neural network models for intrusion detection based on a dataset obtained from real network traffic.

Download a ZIP file on the DLE, called “COMP3003_CW.zip”. This archive contains a dataset, which is extracted from a 10% subset of KDD full dataset (Knowledge Discovery and Data Mining - KDD99 Intrusion Detection Dataset, http://kdd.ics.uci.edu/databases/kddcup99/kddcup99.html). The KDD99 dataset contains 41 features linked with over five million TCP/UDP connection records. Each connection is labeled as either a ‘normal’ (good) connection, or a ‘bad’ connection (called an intrusion or attack). In the dataset, columns 1 – 41 are for features, the 42nd column is for traffic type, either normal, or an attack. For the dataset provided for this coursework, it contains 3,000 connections with 2000 ‘normal’, 500 for ‘neptune’ and 500 for ‘portsweep’, where ‘neptune’ belongs to Denial of Service (DoS) attack, and ‘portsweep’ for probing attack. Two data files are provided as below:

(1). cw_dataset_original.txt (the dataset extracted from KDD99 including symbolic features) (2). cw_dataset_processed.txt (Symbolic features are converted to numerical labels, e.g., the 2nd column, protocol type, tcp=> 1, udp=>2, and icmp=>3; the 3rd column, 32 service types such as ftp and http, are converted to 1 to 32, respectively. Column 42 remains unchanged.

Threshold Criteria (these are indicative only):

< 40% Little or no analysis. Minimum of results, and results are largely incorrect. Little understanding of the subject. Almost no evidence of investigation, evaluation and research on answering the questions. The report is poorly written and structured.

40–49% (Third): Brief discussion and little analysis/investigation. Results are partially correct and/or complete. Little evidence of investigation, evaluation and research on answering the questions. The content and style of the report are mostly appropriate.

50–59% (Lower Second): The majority of the results are correct and complete. Answers are given at an appropriate level of detail and are explained clearly. Some evidence of investigation, evaluation and research on answering the questions. The report is reasonably well structured and the content and style are appropriate.

60 – 69% (Upper Second): A significant majority of the results are correct and complete. Answers are given at great details and are well explained. Clearly and concisely description of how results are obtained. Good evidence of investigation, evaluation and research on answering the questions. The report is of a good standard and structure.

> 70% (First): The results are correct and complete. Especially clear, ambitions and well justified analysis and description. Demonstrating ideas for original thoughts and stretched work. There is strong evidence of investigation, evaluation and research on answering the questions (e.g. deep analysis, fully investigation, good summarise of the investigation). The report is well presented and organised (focused and concisely).