A New Method to Evaluate Subgroup Discovery Algorithms

A Subgroup Discovery algorithms is usually considered better than other method if the average of all its mined subgroups is higher, with respect to some predefined quality measures. This process has some drawbacks: it ignores the redundancy in mined patterns and it might hide important differences among algorithms that return subgroup sets with the same averaged value. In this paper, we propose a new method to evaluate and compare subgroup discovery algorithms. This method starts by removing redundancy using a novel procedure based on the examples covered by the patterns and the statistical redundancy between them. Then, a new similarity and quality methods is used to compared the algorithms based on their ability to detect the patterns and the quality of the mined patterns, respectively. The experimental results obtained show some interesting results that would be unnoticed by the traditional approach.