This document discusses using machine learning techniques for automatic bug triage to reduce the time and costs associated with manually assigning software bugs to developers. It proposes using data reduction techniques like feature selection and instance selection to create a smaller, higher quality bug repository by removing redundant bug reports and words. This reduced dataset would then be used to train a classifier to automatically suggest the most suitable developer for a given new bug, aiming to improve prediction accuracy while reducing training and prediction time compared to using the full dataset.