Ian is the Head of Computer Science and a Reader in Computational Intelligence. He has published 30 papers and has over 630 citations. Research interests focus primarily on data-mining in areas such as crime analysis, house price forecasting and health; soft system problem solving applied to a problems such as allocation of academic staff to teaching, cartographic map generalisation, skeleton feature extraction for the severely disabled, urodynamic pattern matching and container-ship stowage planning; and stereo image processing and tracking in augmented reality.
The potential for using Natural Language Processing in the NHS Wales
The NHS gathers and collates data from many disparate sources. However, much of the data is recorded using Natural Language (that is, relatively unstructured text written using the English language that is dependent on the writing style of the author) or in machine formats, which can be intrinsically hard to interpret. Such sources of data can be easy for someone from a similar background to read and grasp.
Given the above, it seems sensible to investigate the extent to which the use of Natural Language Processing (a sub-field of Artificial Intelligence, which as an area of research is in its ascendency) can be utilised to manipulate these disparate data sources held by NWIS into a format that will enable it to be processed in a meaningful way.
While there has been much research into the development of Natural Language Processing systems there is still much to do. Moreover, when considering data held by NWIS there are additional factors that have to be accounted for. These include the use of medical terminology, the use of abbreviations and frequent cross-referencing.
In addition to the issues associated with Natural Language Processing, research will have to address how to manipulate the original (which is held in various formats) into a structure that can be manipulated easily.
Team members: Laurence Jones, Ian Wilson, Andrew Ware, Penny Holburn
Teaching assignment utilising metaheuristics
Allocation of educators to diverse and rapidly evolving educational programmes of study such as those within Computing and under increasingly tighter budgetary constraints is a non-trivial task given the wide range of subject areas encompassed here, which span information technology, information systems, software engineering and computer science, as well as core areas such as computer forensics and computer games development. Suitability and availability of expertise coupled with an aspiration to limit disruption to existing teaching assignments can often result in first fit solutions that are less than optimal in terms of suitability. This system is highly sensitive to even small changes, which ripple out through assignments and make it a difficult problem for solution. The ongoing work has resulted in a methodology for profiling modules and, by association, educator expertise that provided a basis for exploring a large number of potential teaching assignments utilising search algorithms. The prototype system limited profiles to what staff had previously delivered. The live system saw the integration of data provided by all academic staff within Computing. This data enhanced suitability profiles by indicating which elements of the curriculum staff would and what they could deliver. Implemented in a live format for the first time in preparation for the 2013 academic year, the process rapidly achieved a very good solution to this difficult, evidenced by the significantly lower incidence of change requests from colleagues. The solution minimised disruption to existing assignments, highlighted bottleneck areas and distributed 167 teaching units across 44 members of staff while minimising under and over-assignment despite the very large number of unassigned modules left retiring colleagues. Further work includes the inclusion of more elaborate workload balancing heuristics, the application of other algorithms and expansion into other disciplines.
Team members: Ian Wilson, Ross Davies, Nigel Stanton
Business rule generation for product data quality assurance
This EPSRC-funded project is a partnership with Newport-based company GXS PDQ Ltd. The company produces business rules for customers in the Retail Fast Moving Consumer Goods and Consumer Electronics sectors, and incorporates these rules into its Product Data Quality (PDQ) service for data quality checking. This enables its customers to identify quality failures in their data. However, the generation of business rules is time consuming, and must be repeated following changes in the business environment, the addition of new products or of new channels for sales, or changes to existing products. The project will involve using artificial intelligence (including semantic and statistical data mining techniques) to automate the generation of these rules.
Team members: Ian Wilson, Paul Roach, Valentina Valeva
The use of stereo vision in augmented reality
This work presents a method for non-computationally expensive automatic alignment of cameras that utilises stereoscopic imagery separated at varying distances just below that of the intraocular distance. Here, automatic stereoscopic alignment in real-time is a non-trivial process that relies on calculating the best virtual alignment of camera lenses through image overlaying. This is important as retail 3D camera lenses are not typically sufficiently calibrated for accurate estimates of distance. The alignment of images allows the filtering of background objects and focuses on points of interest. Imprecision in camera lens calibration leads to problems with the required alignment of images and consequent filtering of background objects. The algorithm presented in this paper allows virtual calibration within non-calibrated cameras to provide a real-time filtering of images and the consequent identification of points of interest. The proposed method is capable of generating the best alignment setup at a reasonable computational expense in natural environments with partial background occlusion.
Team members: Ian Wilson, Ross Davies, Andrew Ware
Diagnostic Research into Biomechanical Urinary Problems using Signal Processing
Starting September 2011-2014
Urodynamics is a clinical used to diagnose the pathophysiological reason behind lower urinary tract symptoms with which a patient presents. The test is carried out by taking pressure measurements inside the bladder and rectum and observing how pressure changes during bladder filling and voiding. The data recorded in urodynamics is usually in the form of a time series containing three pressure traces (the two measured pressures, in the bladder and rectum, and the difference between them which is assumed to be bladder muscle activity). The flow rate of any fluid voided and the amount of fluid that has been pumped into the bladder during the test is also recorded. Occasionally X-ray video is used to aid in diagnosis but in most cases the presence of certain pressure events and key values taken during the bladder voiding cycle are used to make a diagnosis. Here, research into the classification of problems derived from trace data is ongoing.
Team members: Ian Wilson, Steve Hogan, Paul Jarvis
Cluster detection and analysis with geo-spatial datasets using a hybrid statistical-hierarchal Neural Network approach
A research project applied to crime data that involved the development of a hybrid predictive model for mapping and visualising crimes and corresponding populations in the study region. GIS was utilised to visualise the location of obtained clusters and burglary incidence 'hotspots’ that were identified using the developed algorithms and associated methodology. Regression analysis was combined with Neural Networks to develop a new hierarchical approach for predicting burglary. The study revealed strong predictors that increased the risk of burglary.
Team members: Ian Wilson, Salar Majeed, Andrew Ware
Developing Image Enhancement and Image Transmission Techniques for an Internet-Based Medical Image Processing System
November 2008 – September 2010
The pointwise sampling algorithm has been used in adaptive contrast enhancement (ACE) for nearly three decades. To overcome its computational overhead various stepwise sampling algorithms have been proposed, including the HOICE (Highly Overlapped Interpolation Contrast Enhancement) algorithm that addresses the artefacts, such as image distortion and blocking effects, of other stepwise sampling algorithms while keeping the advantage of speed. However, HOICE’s internal mechanism has not been compared with traditional pointwise algorithms, which may affect the scope of its future application. This paper establishes a bridge between traditional pointwise sampling and HOICE: it first mathematically analyses the internal mechanism of two typical sampling algorithms (pointwise and HOICE); and then quantitatively demonstrates the similarities between these two kinds of algorithms. Moreover, the results of quantitative measurement suggest that HOICE has an advantage over the pointwise algorithm in reducing the distortion along image edges. Therefore, HOICE has the advantages of both pointwise and other stepwise algorithms making it a reliable sampling algorithm for ACE. Work completed in collaboration with the Medical Physics Department at Heath Hospital, University of Wales has resulted in a novel technique for increasing the contrast in medical images.
Team members: Ian Wilson, Peter Wu, Andrew Ware
A Knowledge Based Engineering System for the Manufacture of Custom Contoured Seating
The aim of this project is to see if it is feasible to produce a Knowledge Based Engineering System (KBES) that will assist a clinical engineer in the manufacture of custom contoured seats for wheelchair users with severe musculoskeletal and postural conditions. The seats created by the KBES should be able to support and accommodate a user’s musculoskeletal and postural conditions.
The KBES will use as an input, the shape of a user’s body which is captured using the Cardiff Body Match (CBM) mechanical shape sensor at the Rehabilitation Engineering Unit (REU) of Cardiff and Vale University Health Board (UHB) and data relating to the user such as age, weight, tissue viability etc. The KBES will then analyse the input data and produce an assessment of the user’s seating ability, from this assessment it is proposed that the KBES will produce engineering rules that dictate the shape of the seat to be produced for this specific user.
The project has three main research areas and they are as follows:
In order to produce the custom contoured seats, engineering rules are required; these will be collected through knowledge elicitation with the clinical engineers at Cardiff and Vale UHB’s REU and through literature review. The second area of research is to design and develop bespoke algorithms that can analyse CBM measurements and produce an output that informs of the user’s musculoskeletal and postural conditions. The algorithms are being created and tested using past CBM measurements of clients with severe musculoskeletal and postural conditions that have been collected in the REU from 1996 onwards. The final research area is the demonstration of a prototype KBES that will produce the custom contoured seats from the inputs.
The project is in collaboration with the REU of Cardiff and Value UHB and is jointly funded by The University of South Wales, Cardiff and Vale UHB and the Engineering and Physical Sciences Research Council (ESPRC).
Team members: Ian Wilson, Adam Partlow and Janusz Kulon
Predicting the geo-temporal variations of crime and disorder
January 2000 – September 2007
We have developed computer models for predicting and mapping crime levels that helped inform the Crime and Disorder Audits of the collaborating Partnerships, in collaboration with local Crime and Disorder Partnerships (South Wales Police, and South Wales Fire & Rescue Service). The work received initial University RS and postdoctoral RA support before EPSRC/CASE funding was secured for two research students.
A system that intelligently interrogates a constantly updated database of crime incidence and provides accurate indicators of where and when crime is likely to be highest would be of great utility in real-time police resource allocation. A limiting factor, however is that crime incidence counts are generally low in relation to crime type, time and space, and subject to randomness. Crime forecast error measures vary inversely with increasing incidence count utilised in estimating time series forecast models. Average crime counts per unit time period and geographic area of at least 25 to 35 are needed before forecast errors become acceptable.
This work details a forecasting framework for short-term, tactical deployment of police resources in which the objective is the identification of areas where the levels of crime are high enough to enable accurate predictive models to be produced. Here, identified hot-spot regions are utilised as the foundation for predictive models.
Team members: Ian Wilson, Andrew Ware, Jonathan Corcoran
Residential property price forecasting
There is no reliable forecasting service for residential values with current house prices taken as the best indicator of future price movement. This approach has failed to predict the periodic market crises or to produce estimates of long-term sustainable value (a recent European Directive could be leading mortgage lenders towards the use of sustainable valuations in preference to the open market value). Work here, underway in collaboration with colleagues, sees the application of sensitivity analysis to artificial neural networks, trained using multivariate time-series data, which forecasts future trends within the housing market. Prior work in this area has been well received both in academic and media circles, with the changing data that encompasses a sustained period of downward prices offering a further opportunity for research. Work on feature selection and data mining led to a predictive model for residential house price forecasting and has resulted in collaborative work in GIS where the impact of flood plains on house prices will be studied.
Team members: Ian Wilson, Andrew Ware
Determining Geographical Causal Relationships through the Development of Spatial Cluster Detection and Feature Selection Techniques
January 2000 – September 2007
In collaboration with the Welsh Consultant Haematologists Leukaemia Registry), this work led to a new algorithm for determining salient attributes within cause and effect relationships. Spatial datasets contain information relating to the locations of incidents of a disease or other phenomena. Appropriate analysis of such datasets can reveal information about the distribution of cases of the phenomena. Areas that contain higher than expected incidence of the phenomena, given the background population, are of particular interest. Such clusters of cases may be affected by external factors. By analysing the locations of potential influences, it may be possible to establish whether a cause and effect relationship is present within the dataset.
Team members: Ian Wilson, Paul Jarvis, Andrew Ware
Container-ship stowage planning
September 1993 – September 2007
A collaborative project with Maritime Computer and Technical Services and (the then) P&O Containers resulted in a paradigm for the automated planning of cargo placement on container ships. The methodology was derived by applying principles of combinatorial optimization and, in particular, the Tabu Search metaheuristic. The methodology progressively refines the placement of containers, using the Tabu search concept of neighbourhoods, within the cargo-space of a container ship until each container is specifically allocated to a stowage location. Heuristic rules are built into objective functions for each stage that enable the combinatorial tree to be explored in an intelligent way, resulting in good, if not optimal, solutions for the problem in a reasonable processing time. This body is heavily cited and forms the basis for a resurgence in interest in this area.
Team members: Ian Wilson, Paul Roach, Andrew Ware