LIDA Seminar Series 2nd April 2020

Resources permitting, we will aim to deliver digitally recorded seminar presentations every Thursday as per the current seminar schedule, although this may not always be possible. Updates on when digital presentations are available to download will made via the LIDA Seminar webpage and the LIDA twitter account (@LIDA_UK).

Follow #LIDAseminar for the latest seminar news.

Presentation 1: Project Update – Advancing analytics for Police resource deployment

By: Rob Long

Abstract: Understanding where and when resources should be deployed is essential to the design of effective crime reduction strategies. The last 15 years has seen a revolution in the use of analytics capable of informing such decisions. A major component of this has been the development of predictive policing analytics – techniques that seek to identify areas at increased risk of volume crime (e.g. burglary, vehicle crime) in the near future. While there have been considerable research efforts to develop these techniques, evaluations of their potential effectiveness in UK policing environments remains limited. To date, predictive policing analytics have concentrated on identifying locations at an increased risk of future crime events. The implicit assumption of these techniques is that all crimes are created equal in terms of prospective resource allocation. However, the practical allocation of crime reduction resources is a considerably more nuanced enterprise which requires practitioners to balance multiple constraints including crime risk, seriousness/harm, and resourcing impact of response.

Building on the ongoing N8 PRP Project – Analytics for Resource Deployment, this project aims to develop new predictive policing analytics capable of incorporating event-weighting schema into spatio-temporal forecasts of crime. These techniques will allow forecasted crime risks to be weighted based on a range of crime and harm reduction priorities. For example, serious offences against the person may be weighted more strongly than property crimes. This approach will also produce multiple forecasts; the second half of the internship will explore how forecasts might be combined using ensemble approaches to solve multiple constraints problems. These efforts will support the design and evaluation of new decision support systems that better reflect the diverse demands of police and their partners.

Presentation 2: Project Update – Changes in meat consumption patterns in the UK – investigation using loyalty card data

By: Patrycja Delong

Abstract: Vegetarian and vegan diets are increasing in popularity in the UK. Main motivations behind these dietary choices include health benefits along with concerns for animal welfare and the environment. Little is known however, on how people’s overall dietary patterns change when they reduce their meat consumption. Traditional food surveys can provide some information about global trends, but lack insight of changes in behaviour at an individual level. Moreover, they suffer from non-response bias and inaccuracy in self-reported food consumption. Using customer transaction data allows us to observe changes in purchasing behaviour of a cohort of households over time.

In this presentation, I am therefore going to be talking about how this project will apply machine learning techniques to identify households that reduce their meat consumption and examine the dietary patterns associated with that transition.

Presentation 3: Project Update – How can we detect Cancer Symptoms from Electronic Health Records?

By: Maab Ibrahim

Abstract: The vast growth in the availability of medical data and the rapid development of data analytics and AI tools and research in the healthcare domain gives us the opportunity, through this project, to facilitate the process of early diagnosis of cancer and cancer reoccurrence identification. We are hoping to do this by helping doctors improve medical decision making to achieve a better healthcare quality through the application of text analytics to Electronic Health Records (EHRs). These records include medical notes that doctors use to describe their patient’s information in unstructured English text. For cancer diagnosis, doctors use EHRs to record the signs and symptoms that are identified as critically related to the potential diagnosis of cancer.

I’ll be talking about how, in this project, we are applying text analytics to EHRs in order to construct a Natural Language Processing (NLP) model which will automatically extract and map NICE cancer symptoms from EHRs. This project is an augmentation of an ongoing project at PinPoint Data Science Ltd on Early Cancer Detection applied on NHS medical data. Not only does this work aim to make cancer diagnosis more efficient and accurate, but it could also impact on reducing costs and expenses in the diagnostic process.

Presentation 4: Project Update – Assessing the presence of food deserts in the UK

By: Francisco Videira

Abstract:Food deserts are vulnerable areas, urban and rural, in which people experience physical and economic limitations in accessing healthy food. Food delivery systems began to appear in the ‘90s for people with reduced mobility who couldn’t go grocery shopping autonomously. In the following years, e-commerce began to unfold to target the wider public, due to the rising improvements in digital technologies. Paradoxically, there is some evidence that a new form of food desert might be emerging: the ‘e-food desert’. These are remote and rural catchments with poor access to physical stores and where food delivery systems are scarce or of poor quality.

By using web scraping techniques, socioeconomic data and spatial interaction modelling, this project aims to assess the presence of these areas and their characteristics, while also trying to reveal ‘hidden e-food deserts’ by simulating changes in availability and cost of delivery services.