PLEASE: A Prototype Learning System Using Genetic Algorithms

Prototypes have been proposed as representation of concepts that are used eeec-tively by humans. Developing computational schemes for generating prototypes from examples , however, has proved to be a dii-cult problem. We present a novel genetic algorithm based prototype learning system, PLEASE, for constructing appropriate prototypes from classiied training instances. After constructing a set of prototypes for each of the possible classes, the class of a new input instance is determined by the nearest prototype to this instance. Attributes are assumed to be ordinal in nature and prototypes are represented as sets of feature-value pairs. A genetic algorithm is used to evolve the number of prototypes per class and their positions on the input space. We present experimental results on a series of artiicial problems of varying complexity. PLEASE performs competitively with several nearest neighbor clas-siication algorithms on the problem set. An analysis of the strengths and weaknesses of the initial version of our system motivates the need for additional operators. The inclusion of these operators substantially improves the performance of the system on particularly diicult problems.