基于Voting机制的IMA-BP不平衡数据分类算法

Classification Algorithm of IMA-BP for Unbalanced Data Based on Voting Mechanism

针对传统分类模型在实际应用中对提取到的不平衡数据特征进行分类时分类结果精度低的问题,提出使用蜉蝣算法(mayfly algorithm,MA)优化的反向传播(back propogation,BP)神经网络分类模型。同时为了提升算法前期全局搜索能力和后期局部搜索能力,引入阻尼比系数和非线性惯性权重因子,构建出改进蜉蝣算法(improved mayfly algorithm,IMA)优化的BP神经网络(IMA-BP)分类器。根据该分类器分类具有随机的特点,引入集成学习中的投票(Voting)机制,将IMA-BP作为弱分类器,将各弱分类器的分类结果通过软投票方法融合,构成了一个Voting机制的IMA-BP分类模型。为验证分类模型的性能,使用UCI数据库中的数据集将该模型与其他的模型进行比较,结果表明Voting机制的IMA-BP分类模型对4个数据集的分类准确率分别为88.67%、96.67%、91.25%、93.52%,都要高于其他模型,说明该分类模型具有较好准确性和可行性,对一些分类任务具有较强的指导作用和应用价值。

Aiming at the problem that the accuracy of classification results is low when the traditional classification model is used to classify the unbalanced data features extracted in practical applications,a BP neural net-work classification model optimized by mayfly algorithm(MA)was proposed.In order to improve the global search ability in the early stage and local search ability in the late stage,damping ratio coefficient and nonlinear inertia weight factor were introduced to construct the improved mayfly algorithm(IMA)optimized BP neural network(IMA-BP)classifier.According to the random feature of the classifier,Voting mechanism in ensemble learning was introduced.Taking IMA-BP as a weak classifier,the classification results of each weak classifier were fused by soft Voting method,and an IMA-BP classification model with voting mechanism was constructed.To verify the performance of the classification model,the model was compared with other models using four datasets from the UCI database.The results show that the classification accuracy of the IMA-BP classification model of the Voting mechanism for the four datasets is 88.67%,96.67%,91.25%and 93.52%,respectively,which is higher than that of other models.It shows that the classification model has good accuracy and feasibility,and has a strong guiding role and application value for some classification tasks.