Mining the MACHO dataset

In order to detect massive compact halo objects (MACHOs) using microlensing, nightly images of the Large and Small Magellanic Clouds and the Galactic Bulge were taken between 1992 and 2000 with a 1.27 meter telescope. The resulting data contains 8·10¹⁰ photometric measurements of star light intensity magnitudes in the red and blue bands for 60 million stars. It has been suggested that the wealth of data may be used to discover new types of variable stars. We briefly outline some methods which may assist the astronomer in classifying variable stars which occur in the MACHO data. First, the almost periodic behavior of many long-period variable stars is used to obtain estimates of the magnitudes on a regular grid and also in regions of missing values. Then some simple features are suggested which characterize the star time series. A classifier based on additive models using these features has been implemented and is part of a tool which can be used in the search for new time series classes.