针对传统的分类算法大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. Aiming at the problem that the traditional classification algorithm mostly aims to minimize the error rate, ignoring the problem of the misclassification type and the imbalance of the data set, a cost-sensitive probabilistic neural network algorithm is proposed. The algorithm introduces the cost-sensitive mechanism into the probability The neural network replaces the misclassification rate by the expectation cost and predicts the minimization of expected cost based on the Bayesian decision rules with the lowest expected cost. The validity of the proposed algorithm is verified by using the field data and data set German Credit The experimental results show that the proposed algorithm has the characteristics of high fault recognition rate, extensive generalization ability and short modeling time.