[[alternative]]The Effect of Feature Scaling on Deep Learning Stock Market Price Prediction

[[abstract]] 近年來機器學習與深度學習模型在巨量資料分析和科技金融方面取得了顯著的成效。時間序列分析主要是利用歷史資料預測未來走勢，然而過去時間序列相關研究較少探討特徵縮放的影響性。本研究利用常見的技術指標，並結合不同特徵縮放及深度學習演算法進行股市價格預測分析。在實證方面利用台灣證券交易所（TWSE）的Α公司2015年到2019年實際股票資料進行驗證，並進行比較分析。綜合上述，本研究目的如下： （1）探討使用不同特徵縮放對於遞迴歸神經網路準確度影響（2）探討加入常見技術指標是否可提高遞迴歸神經網路準確度（3）驗證傳統 ΑRIMΑ 模型與遞迴歸神經網路預測準確度之比較（4）探討不同神經元數及層數對於預測準確度之影響 （5）實際採用2015-2019年Α公司實際股票資料來進行實例驗證[[abstract]] In recent years, machine learning and deep learning models have achieved significant success in huge data analysis and technology finance. Time series analysis is mainly used to predict future trends using historical data, however, the impact of feature scaling has been less explored in past time series-related studies. This study uses common technical indicators and combines different feature scaling and deep learning algorithms to conduct stock market price prediction analysis. Empirical validation is conducted using actual stock data from 2015 to 2019 of Α companies on the Taiwan Stock Exchange (TWSE), and comparative analysis is performed.The purpose of this study is as follows： (1) The effect of using different feature scaling on the accuracy of recursive neural networks.(2) To investigate whether the inclusion of common technical indicators can improve the accuracy of recurrent neural networks.(3) A comparison of the prediction accuracy of the traditional ΑRIMΑ model with that of the recurrent neural network.(4) Investigate the effect of different parameters on prediction accuracy(5) Using an actual stock dataset as an experimental case