Local and global characteristics-based kernel hybridization to increase optimal support vector machine performance for stock market prediction

In this paper, a novel multi-kernel support vector machine (MKSVM) combining global and local characteristics of the input data is proposed. Along with, a parameter tuning approach is developed using the fruit fly optimization (FFO), which is applied to stock market movement direction prediction problem. At first, factor analysis is used for identifying reduced key features called as factor scores from the raw stock index data which when applied to the model contributes to improvement in prediction performance. Subsequently, a hybrid kernel method combining local and global characteristics of input data is proposed, where polynomial is used for global kernel and radial basis function is utilized for local kernel. Additionally, FFO-based parameter tuning scheme is proposed to enhance the prediction performance further. Lastly, the evolving MKSVM with best feature subset and optimal parameters is used to predict stock market movement direction based upon historical data series. For evaluation and illustration purposes, three significant stock databases, NYSE, DJI and S&P 500 are used as testing targets. The effectiveness of this proposed approach is proved by three different stock market datasets, which demonstrate that the proposed approach outperforms the MKSVM with default parameter, MKSVM with PSO, MKSVM with GA and other methods. In addition, our findings reveal that the optimization strategy proposed here may be used as a promising choice forecasting tool for better generalization ability higher forecasting accuracy.