The paper presents an algorithm for finding the value of a discrete attribute which optimal splits a set of examples into two subsets, where the optimality is defined with regard to a criterion which is not computed over individual examples (like entropy) but on pairs of examples. The algorithm reduces the time complexity from quadratic (which we would get with a brute force algorithm) to linear. This is important for qualitative modelling where qualitative change vectors (as used in QUIN) are defined on pairs of examples and not on single examples.
COBISS.SI-ID: 7550548
In the paper, we use one of the methods developed within the project, for construction of preference models. Using the data in which the preferences are given only implicitly, the method can construct a model that can predict the preferences for a concrete example (e.g., a person choosing between a set of alternatives).
COBISS.SI-ID: 7926100
The paper describes methods for induction of qualitative models based on topological analysis and fitting hyper-planes to the data, instead of (explicitly) using linear regression, a method which is described in a separate paper in Artificial Intelligence Journal.
COBISS.SI-ID: 8863572
The paper describes a generalization of derivation to categorical domains. We defined probabilistic qualitative partial derivatives that describe how the change of the variable affects the probability of categorical dependent variable. We propose a local naive bayesian method for computation of such derivatives.
COBISS.SI-ID: 7881812
The paper describes Pade, the crucial method developed within the project. Pade is a new method for qualitative learning which estimates partial derivatives of the target function from training data and uses them to induce qualitative models of the target function. We formulated three methods for computation of derivatives, all based on using linear regression on local neighbourhoods. The methods were empirically tested on artificial and realworld data. We also provide a case study which shows how the developed methods can be used in practice.
COBISS.SI-ID: 8324436