Inductive inference of model structure using hypothesis feedback

The authors describe initial results in the application of methods based on discrete decision theory to the inductive inference of models from collected data. The results are applied to the classification of electric distribution system customers for direct load control. These preliminary results are intended as an exploratory attempt to use discrete decision theory for inference of the structure of large collections of data. The analysis of collected data is modeled as a set of operations which induce equivalence relations on the data and generate meaningful figures of merit for the resulting equivalence classes. The effect is to reduce the data analysis problem to the detection of structure in the figure of merit functions over the set of equivalence classes. An inference algorithm is used to detect this structure and classify the customers into load types.<>