Construction and analysis of data model for financial market volatility prediction based on support vector machine

Financial market volatility prediction is a core issue in modern financial risk management. Traditional econometric methods exhibit limitations when handling nonlinear and high-dimensional data. This study constructs a volatility prediction model based on multi-kernel fused SVM, which adopts an adaptive combination of radial basis kernel and polynomial kernel, combined with a dynamic feature selection mechanism to process complex financial data characteristics. Using 2187 trading days of data from the CSI 300 Index and S&P 500 Index (2015–2023) as samples, we employ a parameter tuning strategy combining grid search and Bayesian optimization to build the volatility prediction model. Empirical results show that the improved SVM model achieves RMSE of 0.0158, reducing by 22.7% compared to the traditional GARCH(1,1) model and 14.6% compared to the basic SVM model. The directional prediction accuracy reaches 68.3%, with only a 7.8% increase in prediction error during high-volatility periods—a significant improvement over traditional models. This model effectively captures the aggregation and nonlinear characteristics of financial market fluctuations, providing crucial data support for investment decisions and risk control.