A Hybrid Machine Learning Approach for Multistep Ahead Future Price Forecasting

Abstract

In the financial sector, prediction of the stock market is one of the imperative working areas. The financial market index price is an important measure of financial development. The objective of this paper is to improve the forecasting accuracy of the closing price of different financial datasets. This work proposes a hybrid machine learning approach incorporating feature extraction methods with baseline learning algorithms to improve the forecasting ability of the baseline algorithm. Support vector regression (SVR) and two faster variants of SVR (least square SVR and proximal SVR) are taken as baseline algorithms. Kernel principal component analysis (KPCA) is introduced here for features extraction. A large set of technical indicators are taken as input features for index future price forecasting. Various performance measures are used to verify the forecasting performance of the hybrid algorithms. Experimental results over eight index future datasets suggest that hybrid prediction models obtained by incorporating KPCA with baseline algorithms reduce the time complexity and improve the forecasting performance of the baseline algorithms.