An empirical study on interpretability indexes through multi-objective evolutionary algorithms

摘要: In the realm of fuzzy systems, interpretability is really appreciated in most applications, but it becomes essential those cases which an intensive human-machine interaction necessary. Accuracy and are often conflicting goals, thus we used multiobjective modeling strategies to look for a good trade-off between them. For assessing interpretability, two different indexes have been taken into account: Average Fired Rules (AFR), estimates how simple comprehension specific rule base is, Logical View Index (LVI), much satisfies logical properties. With aim finding possible relationships AFR LVI, they independent experimental sessions against classification error. Experimental results shown that minimization implies LVI minimization, while opposite not verified.