“Mining” the sky from data mining

Exponential technological advances in telescopes capabilities are today a key factor to obtain large digital sky surveys. The huge amount of digital observational data can be investigated through data-mining techniques based on a clustering approach that uses a number of distinct astronomical objects (such as stars, galaxies, quasar, etc.) as data set, in order to discover rare or even previously unknown types of astronomical objects and phenomena. A very interesting and innovative field in data mining is the outlier detection. Given a set, an outlier is an individual who behaves in an unexpected way or features abnormal properties. We will show an outlier detection technique based on the deviation approach: identification of outliers by examining the spectra of astronomical objects in a group. Objects that “deviate” from an expected description are considered outliers. Also, an algorithm which implements this technique will be described.