A logic-theoretic classifier called Circle

We present a novel classifier based upon principles of logic-theoretic Boolean function minimization. The classifier, called Circle, recursively produces a set of implicants (or rules). The implicant set contains information not only about the presence of features, but also about their absence in determining class values. Thus, Circle´s implicant set is initially non-monotonic with respect to inserting new tuples that have feature values that were not in the training set. One important benefit of this non-monotonicity, however, is that Circle is capable of being robust in the presence of novel feature values. We have created a full implementation of Circle using Java as a host language and Oracle database backend. Because we are interested in data mining in bioinformatics, particularly genomic data, the database was borne out of necessity to both manage and effectively query the information.