Comparison of SVR Techniques for Stock Market Predictions

Abstract

Here we build and test the Support Vector Regression (SVR) model for Indian stock market prediction. The SVR is the regression model based on the primitive Machine Learning (ML) technique called Support Vector Machines (SVM) wherein the sparsity and the parameters of the decision process can be effectively controlled by selecting the desired kernel functions. The conventional ML based prediction and regression models face regularization issues leading to overfitting, unstable learning for nonlinear and multi-dimensional data. We used a strong feature extractor to extricate the parameters indicating the trend of the multi-dimensional financial data. The SVR then correlates the features on a GPU based execution environment for faster predictions. Though our goal is to build a standalone application for the Indian Stock market prediction, as a first step we choose to build and compare SVR models with various kernels to decide whether to buy stocks or not based of the regression model built upon the 20 years of stock market data. The results indicate that the SVR is a very efficient and powerful tool for handling the financial data and can be used in building the stock market predictions.