利用集成學習及離散小波轉換進行股票預測

Abstract

本研究使用台灣上市公司股票之股價資訊、技術指標以及總體經濟指標以集成學習概念進行台灣股市個股漲跌預測、建立最適投資組合。本論文使用五個不同的機器學習模型：決策樹（Decision Tree）、極限梯度提升模型（XGBoost）、輕量化的梯度提升機（LightGBM）、支持向量機（SVM）以及多層感知器（MLP）進行個股的漲跌預測。為了使模型訓練結果更好，本研究利用集成學習（Ensemble Learning）的堆疊技巧（Stacking），將五個機器學習模型的預測結果整合並進行最終的漲跌預測，選出上漲機率較高的股票，接著組成股票投資清單。另外，本研究第二階段使用離散小波轉換（Discrete Wavelet Transform）去除股票收盤價之雜訊，並當作新的特徵加入模型，重新進行預測。實證結果發現，使用多種模型進行集成學習所建立的投資組合能夠獲得更好的績效，且加入小波轉換技術也有效提升模型的整體績效。

This research uses the stock price information, technical indicators, and macroeconomic indicators to predict the trend of individual stocks in the Taiwan stock market with ensemble learning and establish the optimal investment portfolio. This paper uses five different machine learning models: decision tree, XGBoost, LightGBM, SVM, and MLP. To make the model training results better, this study uses the stacking technique of ensemble learning to integrate the prediction results of five machine learning models and selects the stocks with high rising probability, then make up a stock investment list. In addition, in the second stage of this study, Discrete wavelet transform is used to remove the noise of stock closing price, and it is added to the model as a new feature. The empirical results show that the investment portfolio established using multiple models for ensemble learning can achieve better performance, and adding wavelet transform technology can also effectively improve the model`s overall performance.