Research Of Tax Assessment Based On Improved Fuzzy Neural Network

Recently, establishment and maintenance of the tax assessment indicators system is still in the stage of manual operation. The accuracy of tax assessment depends on the officials' judgment and analysis which bring them huge amount of work. Furthermore, the evaluation results are affected by manual factors and not reliable. To improve tax assessment, this paper proposes a tax assessment model based on PSO-FNN-Adaboost. In this proposed model, PSO (Particle Swarm Optimization) is used to optimize FNN (Fuzzy Neural Network) weak classifier, and then Adaboost is utilized to combine multiple PSO-FNN weak classifiers into a strong classifier. The experiment is designed to validate the proposed model. The original model based on PSO-FNN-Adaboost method is trained to get the classification model of the assessment levels. Then the classification model is tested. The experimental results show that the proposed model improved the prediction performance of tax assessment. Compared with single PSO-FNN weak classifier, the accuracy of PSO-FNN-Adaboost strong classifier is increased by 5%.