Stock market forecasting with super-high dimensional time-series data using ConvLSTM, trend sampling, and specialized data augmentation
作者：
Highlights：
• Our model learns high-level features from super-high dimensional time-series data.
• All companies’ price data in the relevant country’s open market are used as input.
• The trend sampling mini-batch sampling method enhances forecasting performance.
• Experimental results show that our model adapts to real-time patterns.
• The model outperforms others with the same training and testing conditions.
摘要
•Our model learns high-level features from super-high dimensional time-series data.•All companies’ price data in the relevant country’s open market are used as input.•The trend sampling mini-batch sampling method enhances forecasting performance.•Experimental results show that our model adapts to real-time patterns.•The model outperforms others with the same training and testing conditions.
论文关键词：Stock market index,Deep learning,Overfitting,Mini-batch sampling,Data augmentation,ConvLSTM
论文评审过程：Received 30 August 2019, Revised 27 June 2020, Accepted 28 June 2020, Available online 8 July 2020, Version of Record 14 July 2020.
论文官网地址：https://doi.org/10.1016/j.eswa.2020.113704
