Learning Qualitative Relations from Categorical Data
Martin Mo, Ivan Bratko · 2010
We address the problem of learning qualitative relations in categorical domains. We propose an algorithm that observes the change of probability of a target class w.r.t. the change in the values of the selected attribute for each learning example. We generalize the notion of a partial derivative by defining the probabilistic discrete qualitative partial derivative (PDQ PD). PDQ PD is a qualitative relation between the target class c and a discrete attribute, given as a sequence of attribute values ai in the order of P(cjai) in a local neighbourhood of the reference point. In a two stage learning process we first compute PDQ PD for all examples in the training data, and then generalize over the entire data set using a machine learning algorithm. The induced model explains the influence of the attribute’s values on the target class in different subspaces of the attribute space.