An approach to inductive learning under errors in data

A generalization of the star-type inductive learning technique is proposed which makes it possible to deal with data with errors whose location and magnitude are unknown, and which are not correctable in practice. Additional knowledge of domain experts is utilized leading to weights assigned to attributes which are aggregated by Saaty´s (1980) analytical hierarchy process (AHP). Minimal length rules containing relevant attributes are sought that cover, for example, almost all of the (positive) examples.