Multicriterial genetic algorithm for pattern generation as a stage of logical analysis of data.

The main stage of logical analysis of data is the generation of patterns that have the maximum coverage of sample observations and also fulfill the constraints on the homogeneity of the coverage. To reduce the effect of overfitting, these constraints should be weakened and the heterogeneity of the pattern should be allowed. However, it is a challenge to determine in advance what kind of heterogeneity is acceptable for each data set. In this paper, we consider the problem of pattern generation as a multicriteria optimization problem where the criteria are the coverage of objects of the target class and coverage of objects that are not included in the target class. To solve this problem, a modified evolutionary algorithm based on the NSGA-II algorithm has been developed. The new algorithm takes into account the representation of the solution (coding patterns) in logical analysis of data and the informativeness of patterns. [ABSTRACT FROM AUTHOR]