A research project, backed by the UK Engineering Physical Research Council between Barclays Bank Plc, Recognition Systems UK Ltd and Aston University, will use neural networking to analyse very large databases and visualise their findings. The research will be undertaken by Prof Chris Bishop of Aston’s neural computing research group and his team, using software from Recognition Systems and data and user input from Barclays Bank. The research council, which seeks to back projects with actual commercial applications, will fund the project to the tune of ú90,000. Managing director of Recognition Paul Gregory explained that the problem faced by large organisations that collect more and more data about customers, is in reading and understanding that data. As co mpanies become ever more customer-focused so their need to understand buying patterns and to tailor their products or services to the individual, increases, he said. Visualisation of data, in graphical or multi-dimensional form, is becoming widely used to enable executives to recognise trends in data quickly and easily, and to act upon them. However, Gregory said that in very large databases, such as Barclays Bank would have, with vast amounts of information on each of its customers, the data becomes noisy. This can mean incomplete or illogical data, in that it defies logical patterns. While traditional analysis techniques cater well for generalisations, organisations increasingly want to look at the exceptions, what individual customers actually do and want. This is where neural computing comes in. The network will be trained on the bank’s data using unsupervised learning techniques. In these, the network is not told what output is required, but is left to run and re-run the data until it learns unusual patterns or trends that traditional techniques might miss. Recognition Systems ran the NeuroData Club with Logica Plc as a part of the UK Department of Trade & Industry’s three-year awareness initiative (CI No 2,589). As a result of this, the company already offers several neural computing software products, including Customer Predictor, which looks at customer trends and attrition rates; Demand Forecaster, which predicts future buying trends; and Data Builder, which cleans and builds missing data with neural networks. But Gregory said none of these adequately handles very large databases and he believes the results of this research project, due in about two years’ time, will benefit all types of organisations worldwide, in areas as diverse as insurance, finance, retailing and government.
This article is from the CBROnline archive: some formatting and images may not be present.
CBR Online legacy content.