“This was a great project at just the right time. It provided a step change in how do business and as a result we've hired two new full-time staff, and sped up the processing and delivery of services.”

Tim MaguireManaging Director, THAT FIGURES

The Project

THAT FIGURES is a company which helps healthcare providers to identify operational inefficiencies in their services. It does so by tracking wireless tags (attached to lanyards) being worn by people (staff and patients) within hospitals over time, and analysing the data gathered from these tags. THAT FIGURES currently uses manual techniques to analyse the data, which limits the scalability of their services and solutions. In order to give THAT FIGURES more opportunity for growth, it developed a project with the Impact Lab to restructure the way it manages, analyses and represents data; enabling THAT FIGURES to more efficiently and accurately identify trends and patterns. The project employed machine learning and data analysis techniques to prove the feasibility of detecting/mapping rooms from data; detecting anomalies in data; identifying patterns of movement on the floor; and highlighting room utilisation and customer wait times.


The Approach

The project was split into four major components, which are summarised as follows:

Mapping room layouts from data

A major part of the room-mapping component involved pre-processing the data. Following this, room layouts were obtained using Seaborn, a Python data visualization library based on Matplotlib, as presented in Figure 1 and Figure 2. Figure 2 explicitly reveals the layout of the healthcare facility:


Figure 1

Scatter plot showing the putative layout of the healthcare facility.

Figure 2

Scatter plot showing the putative layout of the healthcare facility, after further data processing.

Detecting anomalies

This key project component was aimed to identify anomalies in datasets collected by That Figures. The anomalies are particular to drastic changes in positions relative to time (of wireless tags), as seen in the data. In detecting these anomalies, three techniques, namely Standard Deviation (SA), Boxplot Anatomy (BA) and Isolation Forrest (IF), were deployed. SA and BA, which are statistical approaches to detecting anomalies, were used to detect anomalies from the speed of travel of each person wearing a wireless tag. Each wearer’s speed of travel was first deduced from the carrier’s change in position, with respect to the relative change in time. Whereas, IF is a machine learning technique used to detect anomalies from each wearer’s position, along with the time the wearer’s position was recorded. IF performed better than the statistical techniques in detecting seemingly anomalous data.

Mapping patterns of the movement of patients

Each wireless carrier’s entire journey (of the room visits in the healthcare facility) was extracted from the data, after manipulating the dataset. An example journey is shown in the table below. Journeys were then clustered in order to identify patterns, following which a number of patterns were discovered:


An example of a patient’s journey.

Highlighting room utilisation and customer wait times

Patients’ service times (in each of the rooms in the facility), and wait times (to be served in each of these rooms) were deduced from the dataset, as shown in Figure 3 and Figure 4 respectively.


Figure 3

Time spent in each room (in seconds) in total over five days by patients.

Figure 4

Time spent waiting to be served in each room (in seconds) in total over five days by patients.

Figure 5

Time spent in each room (in seconds) in total over five days by members of staff.

The Conclusion

This project has provided That Figures with tools to analyse data in a more scalable and efficient manner. With these tools, they can produce more accurate results, gain valuable insights into operations on the floor, offer bespoke solutions for healthcare services, and ultimately, grow as a business.

The project also serves as a springboard for future collaboration between That Figures and the University of Exeter. A Knowledge Transfer Partnership (between both institutions) is being set up to develop models for automating data analysis, and to employ Machine Learning in identifying opportunities for improving operation.


The Team

That Figures:

  • Tim Maguire, Director

Impact Lab:

  • Dr Ola Oluwasuji, Industrial Research Fellow, University of Exeter
  • Dr Jacqueline Christmas, Senior Lecturer in Computer Science, University of Exeter
  • Kathryn White, Innovation Manager, University of Exeter