Machine learning prediction of stock price behavior in SGX

Abstract

In recent years, ML has been used to solve many complex mathematical problems. Researchers have identified ML as a means to predict stock prices and their characteristics to execute profitable trades in the stock market. This project proposes a novel labelling scheme and evaluates the use of five different ML models, three different feature sets, and two labelling techniques in predicting stock price characteristics within SGX. Our project also examines the tuning of each model and the relations between feature sets and labels. We run our resulting models’ predictions through a backtesting algorithm to evaluate its real-world application. Our experimentation shows that ML predictions can result in profitable trading strategies in the SGX, with the AdaBoost, OHLCV, and our novel threeclass labelling combination offering the highest profitability.