Stock trend prediction using machine learning -- focusing on ensemble method

Abstract

The main purpose of this research is to predict the trend of stock prices using machine learning We apply machine learning classifiers to analyze 16 financial ratios(such as earnings per share return on net worth—after-tax return on net worth—recurring benefits PSR CAPM_Beta etc ) plus four lags of each feature to predict if 250-day moving average of stock price after 4 periods will go up or down The dataset we use includes financial data of common stocks of listed companies and over-the-counter (OTC) companies in all general industries (excluding financial insurance securities) in Taiwan from December 1999 to September 2019 Four of the six classifiers we use are ensemble methods As a result of the implementation the test accuracy of all the four ensemble classifiers is greater than 71% It not only proves that ensemble classifiers perform well but also that we can use the features selected in this study to build a predictive model of stock price trends

Date of Award	2020
Original language	English
Supervisor	Lih-Chyun Shu (Supervisor)