关键词: 少数类过采样, 金融市场极端风险, 粒子群, 最小二乘支持向量机

Abstract:

In order to improve the ability of extreme risk identification and early warning in the financial market, the CSI 300 index is adopted as the research data, and the over-sampling algorithm（SMOTE） of a small number of samples is adopted to solve the problem of sample imbalance. The feature is extracted by factor analysis, and the prediction model（SMOTE-PSO-LSSVM） is constructed by the LSSVM algorithm optimized by PSO. The SMOTE-PSO-LSSVM model is used to predict the CSI 300 index samples from 2007 to 2010. The samples include 193 extreme risk samples, and 154 risk samples are successfully identified by the model, with the recognition accuracy reaching 83.1%. The results show that SMOTE-PSO-LSSVM model has a strong ability to identify financial risk data, can identify risk samples more accurately, has a fast solution speed and high efficiency, and has better performance than traditional BP network and support vector machine. The conclusion of this paper has certain significance for risk identification, market trend control, stock market transaction control and investor decision.

Key words: minority oversampling, extreme risks in financial markets, particle swarm, least squares support vector machines