An architecture for temporal data mining

The work concentrates on a prototype system architecture for mining temporal patterns. As part of that, generic temporal and periodic patterns are formally defined and the mining problem for temporal patterns is described. The components of the architecture are discussed in detail. As part of this architecture, it is recognised that in order to find useful knowledge, the user of a KDD system has to select the relevant data subset, identify suitable classes of patterns and define good criteria for interestingness of the patterns. To help users achieve this, it is expected that powerful languages can be used to express different ad hoc data mining tasks. For this purpose, an SQL-based mining language is presented and it is demonstrated that this language can be used to discover various types of temporal (or not) patterns or rules. (4 pages)