A forecast model for stock trading using support vector machine

The objective of this study is to create a forecast model for the buying and selling points of stocks, using a support vector machine (SVM) model in order to create a highly accurate prediction. The trials compare four Kernel functions of SVM, consisting of Dot function, Radial Basis Function (RBF), Sigmoid function, and Polynomial function to evaluate which Kernel would provide the most accurate forecast results. The trials test the data of five stocks from the Stock Exchange of Thailand. The experiment presents testing and comparing results between using technical data and fundamental data and using algorithms for the selection of data attributes. This study uses the “majority voting” technique to decrease the number of attributes, which uses attribute selection algorithms to screen for attributes that has strong relationships with the fluctuations of stock buying and selling points. The results of the study revealed that the use of the majority voting technique to reduce the number of attributes helps to increase forecast accuracy, and the SVM model using Dot Kernels for data with the selected attributes (majority voting) is more accurate than models using other Kernels.