A Mixed Qualitative-quantitative Classifica't'ion

In this paper a self-learning classification method, able to deal with both qualitative and quantitative data, will be presented. This technique presents important advantages with respect to classical approaches, both numerical and symbolical ones : it is faster, less memory demanding, performs a sequential treatment and can carry out both supervised (non-iterative) and unsupervised learning . Besides, the hybrid connectives based methods yield fuzzy results easy to assess and understand . The first implementation of these classification algorithms, LAMDA, exhibits some interesting capabilities, since the fact of being an inductive classification method with numerical processing abilities places it in a somehow hybrid place . Our technique seems very close to the artificial neural network operation, but presents some advantages over it for classification purposes . This method has been recently implemented as a software tool (offered as shareware), and its capabilities will be illustrated with two examples of process diagnosis, the first one on real data from a pilot plant and the second one on simulated data modeling the behaviour of a communication network. l. LOGICAL CONNECTIVES BASED ALGORITHMS Our proposal to Machine Learning is a kind of inductive techniques that combine some of the best characteristics of both symbolic and numerical algorithms . In order to do so, we rely on the generalizing power of Fuzzy Logic [Zadeh 65], and the interpolation capability of logical hybrid connectives [Piera 87].