Interactive Genetic Algorithms Based on Estimation of User´s Most Satisfactory Individuals

The improvement of the algorithm´s performance and the reduction of a use´s fatigue are important issues in interactive genetic algorithms (IGAs). In order to achieve these purposes, the idea of submitting users the most satisfactory individuals estimated directly is put forward. Firstly, three issues about samples that is a determinative factors of the estimation are discussed, namely when to sample, how to get enough and effective samples and how to evaluate the information that is not included in the samples. Based on the above issues, a method of recognizing users´ most satisfactory and dissatisfactory gene-sense-unit (GSU) and the division of the search space are introduced. Secondly, the method to compose the user-satisfactory individuals with the estimated most-satisfactory GSU is put forward. Thirdly, the computational complexity of the estimation is also given. Fourthly, the tradeoff between user fatigue and sample quality is discussed. The experimental results validated its efficiency. Then the proposed method enriches the method of replacing a user with machine.