Search

Instance selection is a pre-processing technique for machine learning and data mining. The main problem is that previous approaches still suffer from the difficulty to produce effective samples for training classifiers. In recent research, a new sampling technique, called Progressive Border Sampling (PBS), has been proposed to produce a small sample from the original labelled training set by identifying and augmenting border points. However, border sampling on multi-class domains is not a trivial issue. Training sets contain much redundancy and noise in practical applications. In this work, we discuss several issues related to PBS and show that PBS can be used to produce effective samples by removing redundancies and noise from training sets for training classifiers. We compare this new technique with previous instance selection techniques for learning classifiers, especially, for learning Naïve Bayes-like classifiers, on multi-class domains except for one binary case which was for a practical application. [ABSTRACT FROM AUTHOR]

Border identification (BI) was previously proposed to help learning systems focus on the most relevant portion of the training set so as to improve learning accuracy. This paper argues that the traditional BI implementation suffers from a serious limitation: it is only able to identify partial borders. This paper proposes a new BI method called Progressive Border Sampling (PBS), which addresses this limitation by borrowing ideas from recent research on Progressive Sampling. PBS progressively learns optimal borders from the entire training sets by, first, identifying a full border, thus, avoiding the limitation of the traditional BI method, and, second, by incrementing the size of that border until it converges to an optimal sample, which is smaller than the original training set. Since PBS identifies the full border, it is expected to discover more optimal samples than traditional BI. Our experimental results on the selected 30 benchmark datasets from the UCI repository show that, indeed, in the context of classification, PBS is more successful than traditional BI at reducing the size of the training sets and optimizing the accuracy results. [ABSTRACT FROM AUTHOR]