The objective of this work is to present an improved version of a method to learn fuzzy classification rules from data by means of a multi-objective evolutionary algorithm and the iterative approach. The work presented here derives from a preliminary version previously proposed by the authors. In the previous version, the trade-off between accuracy and interpretability during the rule generation process is addressed by defining the accuracy objective, measured by the compatibility of the each rule with the examples and the interpretability objective, defined as the number of conditions in the rule. The best rule to be inserted in the rule base in each iteration is selected among the non dominated solutions, using a criterion related to the accuracy of the rule base. In the new version of the method described here, we propose a new criterion for selecting the best rule, considering the semantic interpretability at the rule base level, specifically the number of fired rules. We also investigate a new form of calculation of the accuracy objective. The experiments show that the new version of the method proposed in this article achieves results that are equivalent to the ones of the previous version with relation to accuracy, although improving both the semantic interpretability at rule base level, evaluated as the number of rules firing at the same time and the complexity at the rule base level, measured as the number of rules and conditions in the rule base.