A peek into the black box: exploring classifiers by randomization

摘要: Classifiers are often opaque and cannot easily be inspected to gain understanding of which factors importance. We propose an efficient iterative algorithm find the attributes dependencies used by any classifier when making predictions. The performance utility is demonstrated on two synthetic 26 real-world datasets, using 15 commonly learning algorithms generate classifiers. empirical investigation shows that novel indeed able groupings interacting exploited different These allow for finding similarities among classifiers a single dataset as well determining extent exploit such interactions in general.