Published online by Cambridge University Press: 26 August 2011
In this paper, we investigate whether incorporating sunspot-groups classification information would further improve the performance of our previous logistic regression based solar flare forecasting method, which uses only line-of-sight photospheric magnetic parameters. A dataset containing 4913 samples from the year 2000 to 2005 is constructed, in which 2721 samples from the year 2000, 2002 and 2004 are used as a training set, and the remaining 2192 samples from the year 2001, 2003 and 2005 are used as a testing set. Experimental results show that sunspot-groups classification combined with total gradient on the strong gradient polarity neutral line achieve the highest forecasting accuracy and thus it testifies sunspot-groups classification does help in solar flare forecasting.