Tax Assessment Based On PNN And K-means Algorithm

Tax assessment is a complex system which is hard to build a accurate mathematical model, furthermore it is a typical pattern recognition problem. The modernization of domestic tax collection and management is relatively low, although a computer information management system has built the establishment of definitely support system is still in the initial stage. The research of tax assessment model which plays a basic role in the revenue management is still in the primary stage.Based on the study of the method of previous in the tax assessment index system, the dissertation propose a way to structure the index system based on Dupont analysis system. After economic significance analysis, a index system of three layers is built, which includs five indexes representing three aspects of the operating conditions. Then select 76 retail taxpayers' financial information from the tax collection and information systems of Cixian Local taxation bureau as the study body. Using the probabilistic neural network to construct the tax assessment model to distinguish the types of the tax payer.In the optimization of the model PNN algorithm is used to distinguish suspicious samples from the training samples. Then construct the collection of suspicious samples by the means of clustering algorithm. The new training samples added a new type called suspicious samples which is different from the compliant and the noncompliant tax payers. Results show that error rate is lower by the model optimized by clustering algorithm, at the same time overcoming the taboo of mistaking the compliant as noncompliant.At the same time the realization method of matlab is studied in the constructing of the model which provides the further software development.