Genetic based concept learning from positive examples

“Version space” proposed by Mitchell (1977) is a typical method of the concept learning from training examples, but this method has some points which can be improved. The purpose of this paper is to construct a flexible learning mechanism which can be applied to the critical points. To do this, the method of concept learning based on genetic algorithm is proposed. The important features of the algorithm are as follows: 1) the system is able to learn the target concept formed by a disjunctive normal form; and 2) if there are some incorrect examples in training examples set, the algorithm will reduce them and generate a correct target concept. This function is called “noise reduction”. Finally, the algorithm is able to learn the target concept from a positive example set. Especially, we note the third feature that is the ability of learning from positive examples