Boosting the Accuracy of Stock Market Prediction using XGBoost and Long Short-Term Memory

Abstract

Stock exchange is one of the famous economical strategy that finally find its way to be experimented with ever-growing Machine Learning (ML) algorithm. With ML, many aspects regarding stock is learnable, to the point where one can predict stock prices. Although tempting, stock price prediction is still a challenging task due to its natural dynamic and real-time movement. Thus, predicting stock prices are deemed unseemingly. On the other hand, different patterns of stock prices are capable of represent a whole lot of detailed data, which is in favor for Deep Learning. In this study, we conducted an experiment of predicting the close stock price for 25 companies. To ensure data reliability and regional notion, these selected companies are officially enlisted in the Indonesia Stock Exchange (IDX). The two ML algorithms used for this experiment are the Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost), both known for its high accuracy of prediction from various representative data. By setting two thresholds, we were able to present a trading approach: when to buy or when to sell. This prediction result from the ML algorithm using in the ensuing trading approach leads to distinct aspects of benefit. In this experiment, XGBoost shown best performance by 99% prediction accuracy result.