探討多種AI模型於台灣股市預測之效能：以SVM、LSTM及集成樹模型為例;Performance Evaluation of AI Models for Taiwan Stock Market Prediction: A Case Study of SVM, LSTM, and Ensemble Tree Methods

NCUIR > School of Management at National Central University > Graduate Institute of Finance > Electronic Thesis & Dissertation > Item 987654321/97887

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/97887

Title:	探討多種AI模型於台灣股市預測之效能：以SVM、LSTM及集成樹模型為例;Performance Evaluation of AI Models for Taiwan Stock Market Prediction: A Case Study of SVM, LSTM, and Ensemble Tree Methods
Authors:	賴韋廷;Lai, Wei-ting
Contributors:	財務金融學系
Keywords:	台灣股市;機器學習;股票預測;長短期記憶網路;集成方法;時間序列預測;Taiwan Stock Market;Machine Learning;Stock Prediction;LSTM;Ensemble Methods;Time Series Forecasting
Date:	2025-07-09
Issue Date:	2025-10-17 12:03:38 (UTC+8)
Publisher:	國立中央大學
Abstract:	近年來，人工智慧(AI)技術在金融市場的應用日益廣泛，其中，機器學習(ML)和一些統計方法被廣泛使用於股票市場的預測。本研究旨在探討如何利用AI模型來預測台灣加權股價指數的變動，並對不同模型的預測效果做比較。我們使用支持向量機(SVM)、長短期記憶網路(LSTM)及決策樹等模型，並使用多種基本面數據，經濟數據和情緒指標作為輸入變數，以增加預測精準度。本研究使用周資料，並使用最大策略虧損(MDD)還有報酬率等指標衡量模型效能。結果顯示，極端隨機樹（ETC）於大多數情境中展現最優異的績效，不僅擁有最高的平均勝率與年化報酬，也在風險控制方面（如最大回撤與夏普值）具有相對穩定表現。本研究顯示AI模型可以有效提升對台股大盤預測能力，未來可以進一步整合強化學習與常識其他輸入變數，以提高預測精準度，並探討在其他標的中的應用潛力。 ;In recent years, artificial intelligence (AI) technologies have been increasingly applied to financial markets, with machine learning (ML) and various statistical methods widely used for stock market prediction. This study aims to explore the use of AI models in forecasting movements of the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) and to compare the predictive performance of different models. We employ models such as Support Vector Machines (SVM), Long Short-Term Memory (LSTM) networks, and Decision Trees, incorporating a variety of fundamental indicators, economic variables, and sentiment indicators as input features to enhance prediction accuracy. The study uses weekly data and evaluates model performance based on Maximum Drawdown (MDD), and return rates. The results show that the Extra Trees Classifier (ETC) consistently outperforms other models across most scenarios, achieving the highest average win rate and annualized return while also maintaining relatively stable performance in terms of risk control (e.g., MDD and Sharpe ratio). This research shows that AI models can effectively enhance predictive capabilities for the Taiwanese stock market. Future studies may further integrate reinforcement learning and incorporate additional relevant input variables to improve forecasting accuracy, as well as explore their potential applications in other financial instruments.
Appears in Collections:	[Graduate Institute of Finance] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	22	View/Open

社群 sharing

Loading...