DTSC301 Applied Machine Learning
Assignment 3
Weighting: 20%
Due Date: 5pm, Friday the 4th of August (end of Week 12)
Submit: your report (PDF) and Jupyter Notebook (.ipynb) via iLearn.
Overview
Conduct a literature review of at least 5 academic articles focusing on the use of RNNs for stock price
prediction (use Google Scholar and focus on the last 5 years). Choosing one of these articles as the
basis of a replication study, build an RNN focused only on using past prices to predict future prices.
Your Literature Review section, focusing on the 5 articles in the very least, should describe for each
of the articles:
a. the data and timeframe used,
b. the metric for success and how well the final mode performed,
c. the RNN structure,
d. any other comments you think are relevant to understanding the results of the paper, including
any other models used for comparison (e.g., Transformers).
A template notebook has been uploaded to iLearn which contains a very simple model and
demonstrates how to download data from the web. You are to use this as a starting point. Your goal is
to build the best model you can. Remember a good model sits right at the border between underfitting
and overfitting – it maximises accuracy while minimizing the network size.
Report Template
Aim to produce an academic journal-style paper that suitably reflects the tasks above. You should
clearly distinguish between the Literature Review and the Modelling. Some guidance has been
uploaded to iLearn. You may adapt the modelling process from your first two assignments.
Focus on explaining why you made the decisions and choices you made. I can see what you did in the
notebook code you upload… what I want to know is why you made those choices.
Marking thoughts
• I value conciseness and elegance in coding.
• I value your explanations as to why you made the choices you made.
• I value the quality of the model – higher test accuracy and a lower number of network
parameters makes a better model. To achieve this takes considerable effort and thought,
which will be rewarded.
• You may want to try and investigate visually what your model has learned and provide some
insights around this.