On the hybridation of artificial intelligence and statistics for effective knowledge discovery in ill-structured domains with messy data

Abstract

Several experiencies highlighted the suitability of combinining AI techniques with Clustering techniques for effective KDD in very complex domains. In this work the benefits of using hybrid methodologies for extracting novel, valid, useful and ultimately understandable knowledge from very complex phenomenons is presented. The importance of including prior expert knowledge as a semmantic biass of the clusters discovery is analyzed as well as the added value of providing interpretation-support tools for assisting both expert and user in the final generation of understandable and explicit knowledge. This approach has shown successful results in some real applications from very different domains. Here results on environmental systems, particularly waste water treatment plants as well as medical domains, spinal cord lesion are presented. We can conclude than classical techniques perform poorly in front of very complex realities, where either algebraic an logics structures have to be modeled to fully explain the domain behaviour. The multidisciplinar approach of designing hybrid methodologies provides very powerful tools to approach those kind of domains.