中文摘要：

准确率和解释性是模糊关联分类模型的两个相互制约的优化目标.目前已有的研究方法中,有的只考虑了分类模型的准确率,有的把模型两个目标转化为单目标问题求解,在模型解释性目标上的优化策略较简单.为此提出一种基于Apriori和NSGA-II多目标进化算法的模糊关联分类模型（MOEA-FACM）,采用基于概率独立性的模糊确认指标筛选生成高质量的模糊关联规则集,以Pittsburgh式的编码方式构建准确率和解释性折中的模糊关联分类模型.标准数据集上的实验表明,该方法所建模型分类准确率比同类模型高,分类模型具有较好的泛化能力,而其所含模糊关联规则的数目和规则前件总的模糊项的个数却较少,模型的解释性较好.

英文摘要：

Accuracy and interpretability are fuzzy associative classification model＇s optimization objective,which complement and restrict each other.So far informed research only takes classification model＇s accuracy into account,or transforms two-objective into single-objective optimization problem.Interpretability＇s optimization method is too simple.In the research field of classification model based on multi-objective optimization and fuzzy rule,most of them generate fuzzy rule according to sample dataset＇s quantitative attribute corresponding fuzzy item＇s permutation and combination.When there are many quantitative attribute in the dataset,evolutionary exploration space is large.So a fuzzy associative classification model based on variant apriori and multi-objective evolutionary algorithm NSGA-II（MOEA-FACM） is proposed.MOEA-FACM adopts fuzzy confirmation measure based on probabilistic dependence to assess fuzzy associative rule in order to generate good quality rule set.Then a small number of fuzzy associative rules are selected from the prescreened candidate rule set using NSGA-II.Maximization of the classification accuracy,minimization of the number of selected rules,and minimization of the total fuzzy items in antecedent of associative rule are regarded as optimization objectives.According to Pittsburgh coding approach and biased mutation operator,a number of non-dominated rule sets,and fuzzy associative classification model,with respect to these three objectives,are built,which can obtain interpretability-accuracy tradeoff.Experiment results on benchmark data sets show that compared with homogeneous classification model,the proposed model has high accuracy,better generalization ability and less number of fuzzy associative rules and total fuzzy items,and better interpretability.