Financial Intelligence Forecasting Model on Regression Analysis and Support Vector Machine

Abstract

The world today is gradually moving into the era of smart economy, and the application of artificial intelligence is bound to trigger huge changes in all walks of life. In the context of the current smart economy, how to use machine learning technology in artificial intelligence to improve the accuracy of enterprise financial analysis has become a hot direction for current research. To address the above issues, this paper proposes a financial intelligence forecasting model based on machine learning models. By establishing a mathematical model to analyse the annual financial statement data published by listed companies, and to determine whether there is fraud according to the model forecasting results. Firstly, from a large number of financial indicators, 60 financial indicators with high frequency of use were selected as variables of the model by using frequency statistics method. Secondly, as financial statement fraud is a typical classification problem, Twin Support Vector Machine (TSVM), a machine learning technique, was chosen and combined with K-Nearest Neighbor (KNN) in order to further improve the forecasting speed and accuracy. In addition, as the data samples for financial statement fraud forecasting are typically unbalanced data, the data are oversampled, undersampled and downsampled in this paper. Finally, for the judgement of model effectiveness, five indicators are selected for analysis in this paper. The experimental results show that compared with other single models, the KNN-TSVM model under the undersampling method has the highest Recall and can effectively identify the fraud samples.