PREDICTION OF STOCK MARKET INDEX MOVEMENT USING PAIRWISE CLASSIFICATION

Abstract

The prediction of index or stock price movements is an attractive and significant research topic for academia and the business world. In recent years, many approaches based on machine learning have been developed to create an effective prediction model. A substantial part of the articles on movement prediction focuses on predicting up-and-down movements of the stock market index and stock prices. This study focuses on four kinds of price movements and proposes a prediction scheme for the emerging multi-class classification task. The proposed approach is mainly based on pairwise classification. The experiments have been conducted on three data sets, namely, the FTSE 100, KOSPI, and S&P 500 indices, using nine technical indicators as inputs. The prediction performance of the approach is compared with the performance of five traditional techniques, multilayer perceptron, support vector machine, naive Bayes, k-nearest neighbor, and regularised multinomial regression. Experimental results based on 11 years of historical data from the FTSE 100, KOSPI, and S&P 500 indices between 2010 and 2021 demonstrate the effectiveness of the proposed pairwise classification-based scheme. The proposed scheme has achieved an accuracy of more than 84%, higher than other techniques. To our knowledge, this study is the first to include the categories presented and to predict the direction of price movements based on such pairwise classification.