Incremental concept formation in an instructional environment

Concept learning continues to be a topic of interest within the fields of computer science and instructional design. The research presented here has implications within both of these fields. The incremental concept formation approach is presented as a viable choice for a general model of concept learning. This approach creates clusterings of hierarchically organized concept categories when presented with previously unclassified instances. Learning is incremental and unsupervised. This research extends previous models of incremental concept formation by presenting an exemplar-based and a probability-based concept learning model. Each model can perform in domains containing nominal, real-valued and mixed data types and can limit the attributes used for classification to those deemed most predictive of class membership. Furthermore, the exemplar-based model uses a global approach to track and correct concept drift. The incremental concept formation approach also has important applications in educational environments. Specifically, when this approach is combined with an algorithm that creates rational sets of matched example/non-example pairs of the concepts to be learned, an environment appropriate for discovery learning is created. An algorithm that creates these rational sets of matched pairs from the concepts contained within a concept taxonomy is introduced. The models presented here are tested extensively in an effort to show their ability to perform well in several situations.