Research on the Fiscal Revenue Prediction Model Based on Data Mining and Grey Neural Networks

The widespread application of data mining technology and the rapid development of machine learning techniques provide a simple and efficient method for predicting local fiscal revenue. The current main model for fiscal revenue prediction involves using data mining techniques for reasonable selection and analysis of data, followed by training neural networks to construct prediction models. This paper proposes a fiscal revenue prediction model based on data mining and grey neural networks, selecting 12 influencing factors that affect fiscal revenue. The Adaptive-Lasso method is used for variable coefficient estimation, and least angle regression is employed to solve the problem, eliminating some variables with lesser impact. The remaining variables are then subjected to GM (1,1) grey prediction to obtain their predicted values, and the prediction accuracy is evaluated with a grading system. Finally, historical data is used to train a BP neural network, constructing a grey neural network combined prediction model, where the grey predicted values are substituted into the trained grey neural network to yield future fiscal revenue predictions. Experimental results indicate that due to the high fault tolerance and adaptability of neural networks, the predicted values fit well with the actual values, with the two curves nearly overlapping. The grey neural network prediction results constructed in this paper are highly reliable.