基于神经网络的微型企业信用评估特征选择及其效果评价

征信机构采集到的所有微型企业信用信息变量并未都适合进行微型企业资信评估,文章设计了一种BP神经网络对此进行特征选择。该BP神经网络的训练基于前向序贯的特征选择算法,以输出层输出对输入值的灵敏度作为特征选择的依据,网络输出最小灵敏度对应的特征变量。通过设计概率神经网络对得到的结果进行仿真分析,信贷机构因此获得的利润比基于列联表分析的特征选择法高2／3。

Feature Selection and its Verification on Micro Enterprise Credit Evaluation:Based on Artificial Neural Network

All characteristics collecting by credit organization are not suitable to be used in micro enterprise credit evaluation, so a BP neural network is designed to select features for this. Training this BP neural network is based on sequential forward selection. The selection criterion is the responsibility of output value of output layer to input value. The network outputs the feature characteristics of corresponding minimum responsibility. A probabilistic neural network is designed to verify its classification ability. The profit of credit institution which used BP neural network to select feature characteristics is increased by 2/3 than crosstabs - based feature selection.