Formation of Probabilistic Concepts through Observations Containing Discrete and Continuous Attributes

The probabilistic concept formation general problem in dealing with mixed-data scale environments is due to the use of different evaluation function for each attribute type. We claim that different behaviors for discrete and continuous evaluation functions are due to an unbalanced contribution for each attribute-type evaluation function inside the main evaluation function. This paper describes an approach based on the difference between the predictability gain for each attribute type. Our approach presents a way to work around for the unbalanced contribution for each attribute-type evaluation functions. Experiments using our approach have shown higher quality in terms of inference ability.