Many real world databases possess time components. Examples range from scientific databases to business databases. Temporal data mining is a technique to deal with problems of knowledge discovery from large temporal databases. In this paper we present a case study of discovering microlensing events from an extremely large astronmical database consisting of 40 million time series (20 million stars with two series each). We discuss the integrated use of techniques from signal processing, pattern recognition, machine learning, statistics and high performance computing to solve this large temporal database mining problem. We take the star light curve classification as a special focus to highlight some temporal data mining issues. We also discuss some computing problems encountered in our work.
展开▼