Investigating the effect of dataset size, metrics sets, and feature selection techniques on software fault prediction problem