Finding kernel function for stock market prediction with support vector regression

Abstract

Stock market prediction is one of the fascinating issues of stock market research. Accurate stock prediction becomes the biggest challenge in investment industry because the distribution of stock data is changing over the time. Time series forcasting, Neural Network (NN) and Support Vector Machine (SVM) are once commonly used for prediction on stock price. In this study, the data mining operation called time series forecasting is implemented. The large amount of stock data collected from Kuala Lumpur Stock Exchange is used for the experiment to test the validity of SVMs regression. SVM is a new machine learning technique with principle of structural minimization risk, which have greater generalization ability and proved success in time series prediction. Two kernel functions namely Radial Basis Function and polynomial are compared for finding the accurate prediction values. Besides that, backpropagation neural network are also used to compare the predictions performance. Several experiments are conducted and some analyses on the experimental results are done. The results show that SVM with polynomial kernels provide a promising alternative tool in KLSE stock market prediction.