Using Lattice-Based Framework as a Tool for Feature Extraction

Abstract

Feature transformation (FT) is one of the way to preprocess data in order to improve classification efficiency. Different FT approaches was intensively studied this past years. These approaches wish to give to a learner only those attributes that are relevant to the target concept. This paper presents a process that extracts a set of new numerical features from the original set of boolean features through the use of an empirical mapping function. This mapping is based on the use of an entropical function to learn knowledge over the Galois semi-lattice construction of the initial set of objects. One advantage here is the reduction of effect of possible irrelevant features. This process allows to design an Instance-based Learning system, IGLUE which uses the Mahanalobis measure. A comparison is done with other ML systems, in terms of classification accuracy and running time on some real-world and artificial datasets.