Student projects

Image

Project supervisor: Gavin Clark 

Project description:

Public Health Scotland (PHS) is Scotland's lead national agency for improving and protecting the health and wellbeing of all of Scotland's people. Their focus is on increasing healthy life expectancy and reducing premature mortality.

One of the key outcomes measured for cancer care is related to how long patients live following their diagnosis - cancer survival. Population-based cancer survival reflects the totality of efforts to improve cancer outcomes from early detection through to prompt diagnosis and effective treatment. This project aims to better understand some of the factors affecting cancer survival in Scotland.

The main factors of interest will be the stage at diagnosis and level of socioeconomic deprivation. The Scottish Cancer Registry (SCR) is a record of cancers diagnosed in those residing in Scotland. The linkage between the SCR and Scottish death registrations allows an understanding of the time from diagnosis to death. The SCR contains information on the stage of diagnosis, as well as postcode information which would allow allocation to deprivation quintiles as defined by the Scottish Index of Multiple Deprivation.

Image

Project Supervisor: Jamie Mariani 

Project Description: 

Trillium Asset Management is an Environmental, Social, and Governance (ESG)-focused investment firm specialising in socially responsible and sustainable investing. 

​​​​​This project covered ESG-focused investment and included statistical analysis and visualization of ESG raw data, ESG scores and stock price performance. This was a terrific opportunity for a curious data scientist to gain first-hand experience in the investment management industry with a global leader in ethical investing. 

A MSc Statistics and Operational Research student worked alongside Trillium's ESG Research Analysts and Portfolio Managers to specify the objectives of the project, identify appropriate sources of data and agreed timelines for completion. The ultimate aim of the internship was to produce a brief report that identifies key ESG patterns, visualises ESG data and shares key ESG findings. Specific tests included: breadth of third-party ESG analysis, distribution of third-party scoring & dynamism of scores, correlation analysis of ESG scores by source, identification of key predictor variables of ESG scores, and performance results of ESG process/scores.

We are committed to research that supports and furthers understanding of fundamental investment and ESG drivers in equity markets. In this context we were delighted to collaborate with a post-graduate student at the University of Edinburgh this summer. Using MSCI data, the statistics student studied ESG’s influence on returns for equity holders.

Jamie Mariani

Image

Project Supervisor: Mark O’Hara (Founder of THRIFT) 

Dates: Session 1

The number of people selling second-hand items online is due to rise from 9 million to 12 million by 2030 in the UK. This project focuses on using technology to help people make more environmentally conscious decisions on how they rehome unwanted items. Ensuring people can easily sell their unwanted or unused items helps reduce the huge amount of waste thrown into landfill each year, and lets people make money from things they no longer need. 

There are many platforms for selling items online (including eBay, Depop, Facebook Marketplace, Etsy). Each of these works in a different way, which can be confusing for potential sellers and makes it harder for people to easily use the services. As a result, people are more likely to throw away unwanted items instead of extending their life. 

The complexity of the re-selling market offers THRIFT an opportunity to simplify things. THRIFT collects data from reselling platforms and uses an algorithm to provide a market price for any item people may wish to sell, streamlining the re-selling process. 

The aim of this project is to use data to help answer the following questions: 

1. What are the key drivers influencing the ‘sold price’ of an item listed on eBay? Can a statistical method be built that takes into account the key drivers for the items in each category and gives the user a market price for an item?
2. Can we assess the relation between the suggested selling price obtained from a statistical algorithm and the actual sold price?
3. For what items / categories listed on eBay does the price at which the item is sold vary over time?

Project Supervisors: Thomas R. Evangelista (Chief Technology Officer, Scrapp LLC), John Scarfo (Chief Information Officer, Scrapp LLC) 

Image

Session: Session 2

Problem

Global waste contamination in recycling, compost, and reuse streams causes massive amounts of otherwise sustainable packaging to be sent to landfill, incineration, or even leaked into the environment. Improper disposal of plastics alone has caused well-known and long-lasting damage to the world’s ecosystems and threatens to seriously harm the global environment. In addition to the clear environmental costs of waste contamination, there is a serious economic factor at play as well: the World Bank estimates that for every 1% reduction in recycling contamination, a large city could save over $1 million annually on waste collection and landfill fees. Compound on top of that that a whopping 83% of global respondents to a PR Newswire study reported that they wanted to know how to recycle better, and it’s clear that now is the time to solve the world’s contamination problem.

What is Scrapp

Scrapp is a purpose-led platform on a mission to reduce global waste contamination, helping to ensure that the world’s waste is properly disposed of in the most environmentally sustainable way possible. We accomplish this by virtually sorting the various pieces of waste scanned through our app, relying on records of the product scanned and the recycling rules of the user’s location.

About the project

The project will focus on using machine learning and AI models to find associations between packaging parts to accurately predict potential packaging information for incomplete product records.

Project aim

The main aim of this project is to use data to help answer the following questions:

1. What are the associations between packaging components for common products?
2. What limitations exist in the development of an AI/ML model for product packaging assessment?
3. How can Scrapp implement AI/ML into the packaging assignment process for new products?
4. Can we produce an automated packaging assessment system using AI/ML to fill in missing or unknown product packaging data?

The insights from the University of Edinburgh’s SWDS Consultancy program have already helped us further understand the data that we need to collect going forward, as well as directly aided our new product packaging assignment process with their insights and solutions.

Thomas R. Evangelista

Being able to listen to how the students are accomplishing the difficult task we have presented them has been a very intriguing experience and I am excited to see the machine learning models they have all created. Throughout the six weeks we have had our eyes opened to the data in ways we have not considered previously.

John Scarfo

Image

Project Supervisors: Alastair Hamilton, Gordon Baggott (Lloyds Banking Group) 

Session: Session 1

Anomaly detection identifies items, events, or observations that show different patterns from other items in the data set. It is widely used in industry such as fraud detection and malicious content removal. Anomaly detection is used in LBG for many of these applications, but cyber and other e-crime threat detection is of particular interest to make the bank most secure for customers and colleagues. 

It is crucial in e-crime threat detection that predictions contain the reliability rate of the estimate to limit false results. By considering how much impact certain inputs have on the output as distributions and using Bayesian Inference and statistical programming techniques, the project can help rank investigations by importance. This project will investigate the most likely threats first. The research will contribute to the continuous evolution of Lloyds Banking Group’s security operations programme. 

The aim of this project is to use data to help answer the following questions: 

1. Can we build a Bayesian Neural Network to find anomalies with confidence intervals in tabular data?  
2. Can we build Bayesian Convolutional Neural Networks that find anomalies with confidence intervals in spatially dependant data such as images? 
3. Can we build Bayesian LSTM networks (long/short-term memory neural networks) that find anomalies with confidence intervals in serialised data such as time-series data?  
4. How do the above networks compare to the results of other anomaly detection algorithms? 

Image

Project Supervisor: Mark O’Hara (Founder of THRIFT) 

Dates: Session 1

The number of people selling second-hand items online is due to rise from 9 million to 12 million by 2030 in the UK. This project focuses on using technology to help people make more environmentally conscious decisions on how they rehome unwanted items. Ensuring people can easily sell their unwanted or unused items helps reduce the huge amount of waste thrown into landfill each year, and lets people make money from things they no longer need. 

There are many platforms for selling items online (including eBay, Depop, Facebook Marketplace, Etsy). Each of these works in a different way, which can be confusing for potential sellers and makes it harder for people to easily use the services. As a result, people are more likely to throw away unwanted items instead of extending their life. 

The complexity of the re-selling market offers THRIFT an opportunity to simplify things. THRIFT collects data from reselling platforms and uses an algorithm to provide a market price for any item people may wish to sell, streamlining the re-selling process. 

The aim of this project is to use data to help answer the following questions: 

1. What are the key drivers influencing the ‘sold price’ of an item listed on eBay? Can a statistical method be built that takes into account the key drivers for the items in each category and gives the user a market price for an item?
2. Can we assess the relation between the suggested selling price obtained from a statistical algorithm and the actual sold price?
3. For what items / categories listed on eBay does the price at which the item is sold vary over time?