Knowledge Discovery With Qualitative Influences and Synergies Jes´ us Cerquides, Ramon L´ opez de M` antaras Artificial Intelligence Research Institute, IIIA Spanish Council for Scientific Research, CSIC 08193, Bellaterra, Barcelona, Spain {cerquide,mantaras}@iiia.csic.es Abstract We review some approaches to qualitative uncertainty and propose a new one based on the idea of Absolute Order of Magnitude. We show that our ideas can be useful for Knowledge Discovery by introducing a derivation of the Naive-Bayes classifier based on them: the Qualitative Bayes Classifier. This classification method keeps Naive-Bayes accuracy while gaining interpretability, so we think it can be useful for the Data Mining step of the Knowledge Discovery process. 1 Introduction Comprehensibility is a key characteristic for algorithm results to be useful in Knowledge Discovery in Databases tasks. Bayesian reasoning has been usually criticized as hard to explain and under- stand, but achieves high performance rates with simple constructs, as happens for instance with the Naive-Bayes classifier[5]. Some approaches to increasing Bayesian reasoning comprehensibility appear in [3,6,12,14,15]. The main idea in all these approaches was to attach linguistic labels as “probable” or “very unlikely” to numerical probabilities, that is to absolutes degrees of belief. Bayesian reasoning works primarily with changes in probability values, and these approaches do not seem to give any interpretation of such changes, giving as result hardly understandable explanations. It has been accepted that, unlike physical parameters, absolute probabilities do not seem to have values (except the endpoints) that are universally interesting [13]. This problem was noticed also by Elsaesser, that in [1] proposed the use of a version of Polya’s “shaded inductive patterns” [10] for linguistic explanation of Bayesian inference. Elsaesser uses Oden’s model [8] to create the linguistic labels related to changes in probability. Elsaesser explanations are comprehensible, but we have no security that reasoning with the information given by these explanations really bring us to coherent conclusions, this is because explanation and reasoning are performed at different levels, and we are not allowed to use a previous explanation in a future case. Another approach is the one followed by Neufeld [7], Wellman [13] and Par- sons [9], using ideas from the field of qualitative uncertainty. The idea behind their work is finding whether a fact A is favoured, unfavoured or not altered by another fact B. Quoting Parsons: