Choosing instance selection method using meta-learning

Many instance selection methods (ISMs) have been widely studied and proposed. But none of these methods obtain good performance on every data set. In this work, we propose an architecture to select the best ISM for a given data set. We use meta-learning to train a meta-classifier that learns the relationship between the ISMs performance and the data set structure. The proposed method was evaluated on public data sets and showed better results than traditional approaches.