Scholarlink

<p>Stock trend prediction, with its huge potential benefits, has attracted considerable attention. However, it is challenging to accurately predict stock trends, not only because the stock market is affected by macroeconomy, market sentiment and other factors, but also because the stock data is characterized by nonlinearity, high noise, and temporality. Also, identifying an appropriate combination of input features to reduce the number of features and improve prediction performance is a goal worthy of effort. In this paper, by using 53 technical indicators as original input features, a stock input feature preprocessing framework combining wavelet denoising and double-layer feature selection (WD) is developed first. Long short-term memory (LSTM) is then used to predict stock trends, and the WD-LSTM model is proposed finally. The experimental results show that the performance of the WD-LSTM model is significantly improved and the number of features is significantly reduced compared with the vanilla LSTM model. Specifically, f1-score on CSI 300, DJIA, NASDAQ, and S&amp;P 500 stock index data sets are increased by 5.5320%, 23.3816%, 45.5732%, and 35.9860%, respectively. Furthermore, the number of input features of those data sets decrease by 79.25%, 69.81%, 71.70% and 71.70%, respectively. The empirical results also indicate that enhancing the quality of input features and determining suitable feature combinations can both improve the model’s prediction performance and mitigate the curse of dimensionality caused by excessive features.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Wavelet Denoising and Double-Layer Feature Selection for Stock Trend Prediction

  • 期刊论文

  • 发表日期：

    2025年3月11日

查看全文
  • Yong Zhang,
  • Jianping Qin,
  • Bocun Lin,
  • Yongbin Su,
  • Xingyu Yang

摘要

Stock trend prediction, with its huge potential benefits, has attracted considerable attention. However, it is challenging to accurately predict stock trends, not only because the stock market is affected by macroeconomy, market sentiment and other factors, but also because the stock data is characterized by nonlinearity, high noise, and temporality. Also, identifying an appropriate combination of input features to reduce the number of features and improve prediction performance is a goal worthy of effort. In this paper, by using 53 technical indicators as original input features, a stock input feature preprocessing framework combining wavelet denoising and double-layer feature selection (WD) is developed first. Long short-term memory (LSTM) is then used to predict stock trends, and the WD-LSTM model is proposed finally. The experimental results show that the performance of the WD-LSTM model is significantly improved and the number of features is significantly reduced compared with the vanilla LSTM model. Specifically, f1-score on CSI 300, DJIA, NASDAQ, and S&P 500 stock index data sets are increased by 5.5320%, 23.3816%, 45.5732%, and 35.9860%, respectively. Furthermore, the number of input features of those data sets decrease by 79.25%, 69.81%, 71.70% and 71.70%, respectively. The empirical results also indicate that enhancing the quality of input features and determining suitable feature combinations can both improve the model’s prediction performance and mitigate the curse of dimensionality caused by excessive features.