Mining for classes and patterns in behavioural data

Abstract

In this paper we compare and contrast the new data mining activity of pattern search with more traditional cluster analysis methods of data mining, in the context of credit data. In particular, we examine a set of behavioural data from a large UK bank relating to the status of current accounts over a twelve month period. We show how conventional clustering approaches can be used, for example to define broad categories of behaviour, whereas pattern search can be used to find small groups of accounts that exhibit distinctive behaviour.