大数据环境下基于限定模糊规则的多分类器

Multi-classifier based on limited fuzzy rules in big data

为解决海量数据环境下,出现分类准确率降低、模型参数训练时间较长等问题,提出一种基于限定模糊规则的多分类器。对确定规则分类过程进行分析和改进,认为确定规则训练过程中后期过于追求精确是导致模型收敛速度下降的关键原因;在此基础上建立总体模型,对收敛程度进行模糊处理,降低模型参数训练时间。通过对确定规则进行正反向规则的补充,减少模糊操作误差,保证分类准确率。通过在Spark计算框架下,与确定规则进行实验对比,验证了该分类器能够有效减少计算迭代次数,提高分类准确率。

To solve the low classification accuracy and long training time of model parameters in big data environment,a multi classifier based on restricted fuzzy rules was presented.The process of the rule classification was recognized and decomposed.It was pointed out that the decline of the model convergence rate was to pursue accuracy completely in the later stage of rule recognition.According to that,the general design model was constructed.Improving the convergence process by fuzzy processing greatly reduced the training time of the model parameters.Adding positive and negative rules to the primordial rules,the error of fuzzy operation was reduced and the accuracy of classification was guaranteed.It was tested on the Spark computing framework and compared with the original methods.It verifies that the proposed classifier can effectively reduce the number of iterations and improve the accuracy of classification.