Knowledge Discovery in Data Mining Highly Multiple Time Series of Astronomical Observations
by
Fuchun Huang
CMA,ANU

This work is motivated by mining the MACHO data of astronomical interests to find the theretically proposed ``dark matter" in physics. Our first concern is the outliers apparently existing in the data set we are mining. We think it is important to know the outliers fisrtly, to try to get more knowledge about them, and to remove them for further analysis. That thinking is not only from the general sense of outliers and general way of data analysing, but also particularly from a preliminary impression on the data that the outliers might come systematically. Therefore we hope to know the time period when outliers came most oftenly, and the reason if possible. To find it out we need very fast algorithm to find outliers out in an irregularly spaced time series which mostly varies periodicly or so. We use robust methods of local fitting parametricly or non-parametricly to find out the outliers. The goodness of the method is evaluated by some theoretical analysis and simulational study.

Date received: July 30, 1999