Imperial College London
My research is focused on a number of closely connected technical areas, developing new methodological tools and applying those tools in a variety of domains. Broadly speaking the technical areas are the analysis of large data sets (data mining, especially anomaly detection), classification, and the analysis of streaming data. Domains in which we are currently applying our new methods include health, finance, and security. Just three illustrations of the work are:
- developing tools for identifying anomalous credit card transactions which might be indicative of fraud.
- the analysis of a large patient safety database, containing records of millions of National Health Service incidents, to detect similar or identical types of incidents with a common cause. This work is in collaboration with the National Patient Safety Agency and is being conducted by a PhD student and myself, and has involved developing techniques to represent very misspelt, partial, abbreviated, and technical descriptions of patient incidents into as points in space so that points (=incidents) which are very close (=similar) can be identified.
- the development of criteria to evaluate the performance of classification rules, so that methods can be compared, optimised, and chosen. In particular, I have developed a radically new method which overcomes a real, but previously unrecognised problem with methods in very widespread use - so that they could lead to sub-optimal methods being chosen. I am in the process of applying these ideas in a variety of application domains, and have already illustrated them in medical diagnosis and the evaluation of credit scorecards used for making decisions about the riskiness of customers.
The potential impact of the various strands of work is considerable - I am currently collaborating with a commercial organisation in applying the ideas in the context of improved investment decisions for major investors, such as pension funds.