代价敏感概率神经网络及其在故障诊断中的应用

针对传统的分类算法人多以误分率最小化为目标,忽略了误分类型之间的差别和数据集的非平衡性的问题,提出代价敏感概率神经网络算法.该算法将代价敏感机制引入概率神经网络,用期望代价取代误分率,以期望代价最小化为目标,基于期望代价最小的贝叶斯决策规则预测新样本类别.采用工业现场数据和数据集German Credit验证了该算法的有效性.实验结果表明,该算法具有故障识别率高、泛化能力强、建模时间短等特点.

Cost-sensitive probabilistic neural network with its application in fault diagnosis

Most original classification algorithms pursue to minimize the error rate, while the differences between types
of misclassification errors and imbalanced dataset are ignored. Therefore, cost-sensitive probabilistic neural network (CS-
PNN) is proposed for the problem in this paper, in which cost-sensitive mechanism is introduced into probabilistic neural
network. CS-PNN is aimed to minimize expected cost, and the misclassification rate is replaced by expectation cost. Class
labels of new instances are predicted by using Bayes decision rules based on minimizing expected cost. The effectiveness
of the algorithm is verified by industrial data and dataset of German Credit. Experimental results show that CS-PNN is
characterized by high recognition for faults, strong ability of generalization and short modeling time.