Financial Data Anomaly Recognition Model Based on Improved Support Vector Machine

Abstract

In order to achieve unsupervised classification of enterprise financial transaction data and identify suspicious abnormal financial data, a financial data anomaly recognition model based on support vector machine is proposed. This article adopts sample testing and independence testing to reduce the dimensionality of the company’s indicators, and selects principal component indicators that can accurately measure the financial condition of the enterprise. When the SVM model is introduced to solve the financial anomaly identification problem, the parameter optimization module are improved. Then, a mixed kernel function through linear combination and optimized parameters using PSO is established. The experiment is performed in MATLAB environment to construct an improved SVM model for feature vector training based on training dataset algorithms. The results indicate that our scheme can detect suspicious values in actual financial account data and effectively avoid overfitting and underfitting phenomena. Compared to similar algorithms, the recognition accuracy and robustness have been improved, making it more suitable for financial crisis warning.