Experimentally studying progressive filtering in presence of input imbalance

Abstract

Progressively Filtering (PF) is a simple categorization technique framed within the local classifier per node approach. In PF, each classifier is entrusted with deciding whether the input in hand can be forwarded or not to its children. A simple way to implement PF consists of unfolding the given taxonomy into pipelines of classifiers. In so doing, each node of the pipeline is a binary classifier able to recognize whether or not an input belongs to the corresponding class. In this chapter, we illustrate and discuss the results obtained by assessing the PF technique, used to perform text categorization. Experiments, on the Reuters Corpus (RCV1- v2) dataset, are focused on the ability of PF to deal with input imbalance. In particular, the baseline is: (i) comparing the results to those calculated resorting to the corresponding flat approach; (ii) calculating the improvement of performance while augmenting the pipeline depth; and (iii) measuring the performance in terms of generalization- /specialization- /misclassification-error and unknown-ratio. Experimental results show that, for the adopted dataset, PF is able to counteract great imbalances between negative and positive examples. We also present and discuss further experiments aimed at assessing TSA, the greedy threshold selection algorithm adopted to perform PF, against a relaxed brute-force algorithm and the most relevant state-of-the-art algorithms. © 2013 Springer-Verlag Berlin Heidelberg.