Research on a Time Series Data Prediction Model Based on Causal Feature Weight Adjustment

• Published in: Applied sciences 2023-10, Vol.13 (19), p.10782
• Main Authors: Huang, Da, Zhang, Qihang, Wen, Zhuoer, Hu, Mingjie, Xu, Weixia
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Artificial intelligence; Banking industry; causal discovery; Causality; Deep learning; Efficient markets; Forecasting; Hypotheses; International relations; Machine learning; Neural networks; Regression analysis; Securities markets; Stock exchanges; Stock markets; stock price forecast; Stock prices; Support vector machines; Time series; time series data; Weather forecasting
• ISSN: 2076-3417; 2076-3417
• DOI: 10.3390/app131910782

As the Information Age brings people an amount of data, research on data prediction has been widely concerned. Time series data, a sequence of data points collected over an interval of time, are very common in many areas such as weather forecasting, stock markets, and so on. Thus, research on time series data prediction is of great significance. Traditional prediction methods are usually based on correlations between data points, but correlations do not always reflect the relationship exactly within the data. In this paper, we propose the LiNGAM Weight Adjust–LSTM (LWA-LSTM) algorithm, which combines a causality discovery algorithm, feature weight adjustment, and a deep neural network for time series data prediction. Several stocks in the Chinese stock market were selected and the algorithm was used to predict the stock price. Comparing the prediction effect of the model with that of the LSTM model alone, the results show that the LWA-LSTM model can find the stable relationship between the data better and has fewer prediction errors.