Tax Credit Risk Classification Based On Adaptive BP Neural Network

The tax credit is an important part of the social credit system, the integrity of the tax reflects the social credit situation in a certain extent. Tax credit risk classification management, its essence is that different levels of credit risk taxpayers are classified, depending on the circumstances of the taxpayer to fulfill tax obligations, and to carry out targeted and differentiated management according to the different taxpayer credit risk. The rapid development of the tax informationization provides a lot of data support for the tax credit risk management, how based on historical tax related data, using data mining technology to realize the tax credit risk scientifically valid identification and classification is a new subject in the current practice of the tax department tax source management.Firstly, considering the shortcoming of BP neural network applying in the process of tax credit risk classification, such as slow convergence, high classification accuracy and generalization capability shortcomings of poor. Tax credit classification algorithm based on BP neural network error function optimization is proposed. The algorithm integrates the compound error function into the traditional BP neural network algorithm and introduces the concept of the hidden layer error function. The weight of BP neural network is adjusted according to the rate of change of the error and using the error rate of change as the weight. In the later stage of the model training, the algorithm also can speed up the convergence speed of the network and avoid falling into local minimum depending on a large error rate of change, this provides the guarantee for the training of tax credit risk classification model.Secondly, in the process of constructing a tax credit risk classification model based on BP neural network algorithm, in order to solve the problem of performance defect caused by the fixed learning rate selection in the traditional neural network. Tax credit classification algorithm based on BP neural network with hierarchical adaptive mechanism is proposed. The algorithm was implied layer and output layer learning rate setting and can adaptively adjust learning rate according to the size of the error and error trends, it is for ensuring adaptively learning rate is maintained at a relatively modest position, and in order to speed up network convergence and increase credit risk classification model classification accuracy and stability.Finally, two algorithms aboved are experimentally verified, the experimental results show that compared with the traditional BP neural network algorithm, BP-OEF algorithm which is based on BP neural network error function optimization; and BP-LAAM algorithm which is based on BP neural netwo rk with hierarchical adaptive mechanism generating tax credit risk classification model has faster convergence rate, higher precision and stronger generalization ability. The experiment demonstrate s the effectiveness of the algorithm.