使用MFCC特徵提取技術結合LSTM 模型之股價預測—以特斯拉股價為例;Tesla Stock Price Prediction through MFCC Feature Extraction and LSTM Modeling

NCUIR > School of Management at National Central University > Graduate Institute of Business Administration > Electronic Thesis & Dissertation > Item 987654321/94864

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/94864

Title:	使用MFCC特徵提取技術結合LSTM 模型之股價預測—以特斯拉股價為例;Tesla Stock Price Prediction through MFCC Feature Extraction and LSTM Modeling
Authors:	洪子鈞;Hong, Tzu-Chun
Contributors:	企業管理學系
Keywords:	股價預測;梅爾頻率倒譜係數;長短記憶網路;支持向量機;Stock Price Prediction;Mel Frequency Cepstral Coefficients;LSTM;SVM
Date:	2024-07-11
Issue Date:	2024-10-09 15:34:37 (UTC+8)
Publisher:	國立中央大學
Abstract:	過去十幾年間，全球經濟快速成長，促進了金融業的繁榮發展。隨著生活水平的提高，股票市場不僅是企業籌集資金的主要途徑之一，也成為大眾普遍採用的投資方式。然而，由於股票市場具有隨機和非線性的特性，使得股價預測變得非常具有挑戰性。過往研究中，學者們已經探索了多種數值數據和文本數據的方法來預測股價。受此啟發，本研究提出了一種基於梅爾頻率倒譜係數（MFCC）的特徵提取方法，將股價歷史數據轉換為波形數據，並提取多維度的特徵向量，以提供一種新的股價預測方法，期望在準確性和可靠性方面取得顯著提升。本研究透過將MFCC 提取的特徵數據集切割為短期、中期和長期數據，以符合股價技術線圖中的周線、季線和月線趨勢，從而有助於模型在不同時間粒度下的訓練特徵，輸入LSTM 模型與SVM 模型進行訓練。最終，本研究的結果證明使用MFCC 提取多維特徵數據結合LSTM 模型，在股價預測領域具有一定的準確率，為未來的研究提供了一種新的特徵提取方法和方向。;In the past decade, the global economy has grown rapidly, promoting the prosperity of the financial industry. With the improvement of living standards, the stock market has not only become one of the main ways for companies to raise funds but also a common investment method for the public. However, due to the random and nonlinear nature of the stock market, stock price prediction has become very challenging. In previous studies, scholars have explored various methods of numerical data and text data to predict stock prices. Inspired by this, this study proposes a feature extraction method based on Mel Frequency Cepstral Coefficients (MFCC), converting historical stock price data into waveform data and extracting multi-dimensional feature vectors to provide a new method of stock price prediction, aiming to achieve significant improvement in accuracy and reliability. This study cuts the feature data set extracted by MFCC into short-term, medium-term, and long-term data to match the weekly, monthly, and quarterly trends in stock price technical charts, thereby assisting the model in training features at different time granularities, and inputs the LSTM model and SVM model for training. Ultimately, the results of this study demonstrate that using MFCC to extract multi-dimensional feature data combined with LSTM models has a certain accuracy in the field of stock price prediction, providing a new feature extraction method and direction for future research.
Appears in Collections:	[Graduate Institute of Business Administration] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	53	View/Open

社群 sharing

Loading...