Further evaluation of pruning in learning boolean functions

This work continues an earlier analysis (Castineira and Monard, 1990; Nicoletti and Monard, 1993) of the problem of pruning, within a framework of automated feature construction when learning boolean functions. Automated feature construction is implemented through three different biases, namely root, fringe and root-fringe. It presents an empirical evaluation of two pruning techniques (reduced error pruning and of its variation) based on their application to trees generated through an automated feature construction. These techniques, although at first studied only for classical boolean functions, appear very promising for an analysis of fuzzy boolean connectives