Machine Learning
Department of Computer & Information Sciences
LEARNING OUTCOMES
The learning outcomes (LOs) for this module are:-
Knowledge & Understanding
LO1 Demonstrate knowledge and understanding of the core concepts of machine learning
and its underlying mathematical foundations
LO2 Demonstrate knowledge and understanding of the principal advanced machine
learning techniques for solving real world problems.
Intellectual / Professional skills & abilities
LO3 Critically evaluate machine learning algorithms and applications.
| ASSESSMENT BRIEF | |
| Module Title: | Machine Learning |
| Module Code: | KV7006 |
| Academic Year / Semester: | 2021-22 / Semester 2 |
| Module Tutor / Email (all queries): | Ossama Alshabrawy [email protected] |
| % Weighting (to overall module): | 60% |
| Assessment Title: | Design, Development, Analysis and Performance Evaluation of Deep Learning algorithms |
| Date of Handout to Students: | |
| Mechanism for Handout: | Module Blackboard Site & Online session |
| Deadline for Attempt Submission by Students: |
25th May 2022 at 23:59 |
| Mechanism for Submission: | All the code files (python files or jupyter notebook) along with the report should be compressed (.zip format) and uploaded to Module Blackboard Site |
| Submission Format / Word Count | The word limit for the report is 1600-2000 words, not including the front cover, table of contents page, references and appendices. Please upload your files as a zip format only. |
| Date by which Work, Feedback and Marks will be returned: |
22nd June 2022 |
| Mechanism for return of Feedback and Marks: |
Mark and individual written feedback will be uploaded to the Module Site on Blackboard. For further queries please email module tutor. |
AssignmentTutorOnline
LO4 Analyse, design and develop machine learning solutions and evaluate their
performance
Personal Values Attributes (Global / Cultural awareness, Ethics, Curiosity) (PVA)
LO5 Carry out independent research, individually and a part of a team, and communicate
effectively the research findings.
This assessment addresses learning outcomes LO1, LO2, LO3 and LO4.
Nature of the submission required:
The report and supporting code should be attached and submitted in a zip file. The zip file should
identify student by name and ID. Only .zip files will be accepted.
Instructions to students:
This is an individual piece of work, and you must not work with others to construct your work. During
the semester there are numerous opportunities to seek and get advice and support on your work,
from tutors and peers but you must ensure you do not do work for others or copy work from others.
Academic Conduct:
You must adhere to the university regulations on academic conduct. Formal inquiry proceedings will
be investigated if there is any suspicion of misconduct or plagiarism in your work. Refer to the
University’s regulations on assessment if you are unclear as to the meaning of these terms. The latest
copy is available on the university website.
https://www.northumbria.ac.uk/about-us/university-services/academic-registry/quality-andteaching-excellence/assessment/guidance-for-students/
If you need an extension:
Contact ask4Help. Tutors and Module tutors cannot grant extensions.
Make sure that your report is submitted on time. University regulations state that assignments
submitted late without approval will incur a 10% reduction for the first 24hours then a zero mark after
this.
You may apply for an extension of time to complete assessed coursework if there are personal
circumstances which are unforeseen and unpreventable and have a serious effect on your ability to
submit the work by the published hand-in deadline. You must submit an ‘Application for Authorisation
for Late Submission of Assessed Work’ before the hand-in deadline. Appropriate medical certification,
or other relevant evidence confirming the circumstances, must be provided. Information regarding
this policy and procedure can be accessed below:
https://northumbria-cdn.azureedge.net/-/media/corporate-website/new-sitecoregallery/services/academic-registry/documents/qte/assessment/guidance-for-students/pl,-d-,008-
v004-late-submission-of-work-and-extension-requests-policy.pdf?modified=20200803152930
Disabled students
Contact the module lead tutor about reasonable adjustments.
Errors
If any errors are found in this document, changes will be posted to the eLP (Blackboard). Versions will
be clearly stated. All versions will be accepted.
INSTRUCTION OF ASSESSMENTS
Assessment Brief
Lately, Deep Learning has had a tremendous amount of attention especially in medical image
analysis. In this assignment you will be required to design, develop, analyse and evaluate an
appropriate deep learning model. You can build your own model or use a pretrained model with
your layers added to it. You will explore the dataset and then apply that model to a dataset of your
choosing. You will need to evaluate the performance in terms of precision, recall, F1-score, ROCcurve and PR-curve. You will need to compare models to add more value to the analysis. You will
discuss the findings that have been produced, and critically reflect upon the model and its
predictions.
Assessment Tasks:
You have been provided with access to four datasets; all are available on Kaggle (Please see links
below). The data covers the following scenarios:
• Cell images for detecting Malaria
• ECG Heartbeat classification
• Classification of breast cancer images
You are required to choose ONLY one of the above scenarios as your assignment. Your task is to
produce a deep learning model that is appropriate to the problem. The model can be your own
model or designed based on fine-tuning of a pretrained model. You are required to conduct data
preparation/transformation to make the data ready for the model. Please note that what will
be provided in the report should reflect on the python code. Please also note NOT to take
on any existing code online as your own work. The errors in the code will affect your final
mark. The key components you must complete are:
1. Explore the dataset to understand its characteristics [10 Marks]
2. Pre-process your data to be suitable for building the model [10 Marks]
3. Build the models that allows for the task specified for chosen dataset and that are going
to be used in your comparisons [20 Marks]
4. Evaluate the models’ predictions using the metrics stated above. [15 Marks]
5. Fine-tune the best model to get better predictions on the test set [15 Marks]
6. Present your findings with suitable visualisations that are easy to interpret [15 Marks]
7. Critically evaluate and discuss the whole process and the findings and what can be
improved [15 Marks]
Please note: to ensure fairness across the different datasets/research questions, the actual performance (e.g.
measured through metrics such as accuracy) of your model will not contribute towards the mark that is
received.
Datasets:
1. Malaria cell images dataset: https://www.kaggle.com/iarunava/cell-images-for-detectingmalaria
2. ECG Heartbeat categorization dataset: https://www.kaggle.com/shayanfazeli/heartbeat
3. Breast histology images to classify invasive ductal carcinoma (IDC):
https://www.kaggle.com/simjeg/lymphoma-subtype-classification-fl-vs-cll
APPENDIX A
Marking criteria
| Grade | Criteria |
| 70 – 100 % |
A mark of 70% or over is indicative of excellent work where the student has more than met the requirements of the assessment brief and demonstrated an exceptional understanding of deep learning models, tools and techniques along with knowledge of their chosen dataset and provides a comprehensive critical view of the workflow of these models and excellent presentation of the results by distinctive visualisations. |
| 60 – 69 % |
A mark within this range is highly competent and completed to a high standard. The work demonstrates a good level of understanding of deep learning models along with knowledge of their chosen dataset and provides a comprehensive critical view of the workflow of these models and good presentation of the results by visualisations. The requirements of the assessment brief have been met to a high standard but with room for a few minor areas of improvement. Marks at the lower end in this band suggest that students have met all or most of the requirements of the assessment brief but there are a larger number of minor areas needing improvement. |
| 50 – 59 % |
A mark within the range indicates a pass, where the work has been completed to a satisfactory standard, but where there is still significant scope for improvement. The work demonstrates an acceptable understanding of deep learning tools and techniques along with a reasonable knowledge of their chosen dataset and provides a reasonably well-documented account of the workflow of these models. The work will have covered most of the key assessment criteria, but these might be at a more superficial level compared with work in the higher mark ranges, with evidence of a less complete understanding of the subject area. The work may indicate that less independent learning has been performed or that less robust methods are used. |
| 40 – 59 % |
This indicates a fail mark, where learning outcomes may not all have been met to a satisfactory standard and where there may be a range of omissions, poor communication and/or possibly a lack of knowledge derived from wider reading. The work does not demonstrate an acceptable understanding of deep learning models, nor provides a well-documented account of workflow of these models. Work in this mark range indicates insufficient evidence of an understanding of the subject area appropriate to level 7, and/or that insufficient attention has been given to the assessment brief. |
| 15 – 39 % |
This indicate a fail mark, indicating a piece of work which is below the acceptable standard and which provides little evidence of the skills, understanding or knowledge appropriate to level 7. There may be many errors and omissions, and few or no of the learning outcomes have been |
| met, with an inadequate demonstration of the knowledge required of key deep learning models/tools and techniques and their applications. Instructions may not have been followed or assessment criteria may have been missed out. |
|
| 0 – 14 % |
missing mostly irrelevant. |