| EPSRC Reference: |
EP/C532589/1 |
| Title: |
Statistical and machine learning tools for plastic card and other personal banking fraud detection |
| Principal Investigator: |
Hand, Professor D |
| Other Investigators: |
|
| Researcher Co-Investigators: |
|
| Project Partners: |
|
| Department: |
Mathematics |
| Organisation: |
Imperial College London |
| Scheme: |
Standard Research (Pre-FEC) |
| Starts: |
01 October 2005 |
Ends: |
31 March 2008 |
Value (£): |
233,935
|
| EPSRC Research Topic Classifications: |
| Artificial Intelligence |
Image & Vision Computing |
| Information & Knowledge Mgmt |
|
|
| EPSRC Industrial Sector Classifications: |
|
| Related Grants: |
|
| Panel History: |
|
|
Summary on Grant Application Form |
|
Fraud is a major problem in the personal banking sector. It is tackled by prevention and detection, but it is constantly evolving as fraudsters find new ways through existing protective barriers. This project aims to devise new methods and algorithms for detecting transaction profiles characteristic of fraudulent activity. It is based on our previous work on 'pattern detection and discovery': the discovery of rare anomalous configurations in a database.The project has two components:(i) Databases of previous transactions provided by our collaborating banks (which include Capital One, Abbey, Lloyds TSB, and Alliance & Leicester), will be modelled to provide a broad description of typical transactions patterns. Using these models, we identify anomalous behaviour, such as transaction patterns which are very unusual, or transaction patterns which are surprisingly similar to each other, using 'pattern discovery' methods. Pattern discovery is a relatively new subdiscipline of data mining, concerned with detecting peculiarities in databases, rather than modelling them in their entirety.(ii) A database of transaction patterns which have previously been known to be fraudulent, will be used to provide seeds for learning and extrapolation algorithms. These are algorithms which automatically generalise from the patterns presented to them. Using these tools, we apply automatic machine learning and statistical methods to extrapolate from existing fraud patterns, as well as from those newly discovered in part (1). Here we again use pattern discovery methods to try to find common features amongst these fraudulent patterns.Patterns may be anomalous relative to previous usage of an account by the same customer, relative to the broad population of other customers, or relative to a 'peer group' of customers who previously behaved in a manner similar to a given customer. One of the key aspects of the work will be exploration of different ways of defining 'dissimilarity' between transaction patterns.The algorithms will be developed at both individual transaction level and account level, and a key aspect is that the methods should be applicable as new data (transactions) arrive: the sooner an anomaly or fraud is detected, the better. However, one of the challenges is to avoid making too many false positives (saying a transaction is fraudulent, when it is not). This is particularly problematic, because typically less than I in a thousand transactions are fraudulent in retail banking.
|
|
Key Findings |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Potential use in non-academic contexts |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Impacts |
|
| Description |
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk |
| Summary |
|
| Date Materialised |
|
|
|
|
Sectors submitted by the Researcher |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
| Project URL: |
|
| Further Information: |
|
| Organisation Website: |
http://www.imperial.ac.uk |