基於循環神經網路與群聚分析之最佳網路結構及股價分析

Optimal Structure And Stock Price Analysis Base On Recurrent Neural Network And Similarity Clustering

顏劭聿(SHAO-YU YAN)

指導教授 ： 劉寅春

淡江大學/工學院/電機工程學系碩士班/碩士(2022年)

https://doi.org/10.6846/TKU.2022.00267

摘要

對於時間序列數據的研究及其預測中，股市走勢預測是最具代表性的一部分。準確預測股票為市場交易者帶來更多利潤並有助於財務決策製作。有各種機器學習以及深度學習模型有助於預測股市的精準度。本篇研究的主要目的為發掘一個最佳的模型架構，並且可以應用在預測股票S P500指數價格的變化。本文提出的方法強調數據的預處理，大量論文對於個股的分析較為重視，而忽略了同類股票之間的相關性。本研究提出了一種挖掘相似股票的聚類方法，根據對於S P500中的所有股票進行分析。分析的項目包含（歷史的走勢，不同價位的天數等），最後繪製出所有的500隻股票之風險-收益分佈圖。並查找不同股票之間的相似度，繪製出相似度圖，最終用Heatmap呈現結果，用更加直觀的方式最快速的尋找。為了時間序列模型的預測結果RMSE更佳，本研究通過改變不同參數於尋找的不同單元一個最好的模型，可以用來預測未來的股票價格。評估是通過使用開源的股票市場可訪問數據集，具有開盤價、最高價、最低價和收盤價。

關鍵字

循環神經網路 ； 群聚分析

並列摘要

In the research and forecasting of time series data, the stock market trend forecast is the most representative part. Accurately predicting stocks brings more profits to market traders and aids in financial decision making. There are various machine learning and deep learning models that help predict the accuracy of the stock market. The main purpose of this research is to discover an optimal model architecture that can be applied to predict the price changes of the stock S P500 index. The method proposed in this paper emphasizes data preprocessing, and a large number of papers pay more attention to the analysis of individual stocks, while ignoring the correlation between similar stocks. This study proposes a clustering method for mining similar stocks, based on the analysis of all stocks in the S P500. The items analyzed include (historical trends, days at different price levels, days in the red, days in the black, risk and return), and finally draw the risk-return distribution map of all 500 stocks. And find the similarity between different stocks, draw a similarity map. Finally, use the Heatmap to present the results, and find the fastest in a more intuitive way. In order to make the RMSE of the prediction better. We change the epochs, batch, the number of hidden layers, and the number of neurons in different hidden layers to find a best model for different units, which can be used to predict the future stock price. The assessment is made by using open source stock market accessible datasets with open prices, high, low and close prices.

並列關鍵字

Recurrent Neural Network ； Similarity Clustering

參考文獻

參考文獻

[1] Md. Arif Istiake Sunny, Mirza Mohd Shahriar Maswood and Abdullah G. Alharbi, "Deep Learning-Based Stock Price Prediction Using LSTM and Bi-Directional LSTM Model," in 2nd Novel Intelligent and Leading Emerging Sciences Conference, NILES, 2020.

[2] Sepp Hochreiter and Jürgen Schmidhuber, "Long Short-Term Memory," Neural Computation, vol. 9, no. 8, pp. 1735-1780, 15 November 1997.

[3] Ferdiansyah, Siti Hajar Othman, Raja Zahilah Raja Md Radzi, Deris Stiawan, Yoppy Sazaki and Usman Ependi, "A LSTM-Method for Bitcoin Price Prediction: A Case Study Yahoo Finance Stock Market," in International Conference on Electrical Engineering and Computer Science (ICECOS), 2019.

[4] Raghavendra Kumar, Pardeep Kumar and Yugal Kumar, "Analysis of Financial Time Series Forecasting using Deep Learning Model," in 11th International Conference on Cloud Computing, Data Science Engineering (Confluence), 2021.