Selected for Regular Presentation of Prestigious ‘IEEE BigData 2022’ International Conference and Received Student Travel Award

▲ Seo Kang-hyeon from the Integrated Ph.D. Program and Professor Yang Ji-hoon (from left)

It was confirmed that the paper titled Exploring Multi-Time Context Vector and Randomness for Stock Movement Prediction, written by Seo Kang-hyeon (first author) from the Integrated Ph.D. Program and Professor Yang Ji-hoon (adviser, corresponding author) from the Machine Learning Lab of the Department of Computer Science and Engineering, was published as one of the regular papers of the “2022 IEEE International Conference on BigData” (IEEE BigData 2022). Additionally, Seo Kang-hyeon, the paper’s first author, received the Student Travel Award at the conference.

Listed as one of the excellent software conferences by the Korean Institute of Information Scientists and Engineers, the IEEE BigData is an international conference that presents and shares recent theories and various examples of applied research in the field of computer science. Moreover, the Student Travel Award, which will be presented to Seo Kang-hyeon, is given out to a selected student in recognition of their outstanding paper, along with a small amount of money for participation.

This research pertains to predicting the increase or decrease of stock-market closing prices the next day compared to the day before using the method of binary classification. The research applied the recurrent neural network that shows excellent performance in learning time-series data and the multi-head attention method that is frequently employed in the areas of natural language processing and computer vision.

The research team combined the two deep learning models mentioned above after focusing on the fact that people observe candlestick charts at different lengths to gain information for their prediction about the next day’s stock-price movements before deciding on an investment. In addition, the research set the trainable Gaussian noise distribution and sampled noise from the distribution to use in the prediction of stock prices to apply the stock-price movement by a “Noise Trader (a trader who blindly buys and sells stocks based on subjective judgments and groundless rumors, not on rational analysis and judgments with correct information),” which was mentioned in the market microstructure as one of the fields of economics.

The experimental results of this research showed that its prediction of stock prices did more to improve prediction accuracy and reliability than the latest model at that time. It also demonstrated more cumulative returns in the stock-based portfolio trading simulation using backtesting.

▶IEEE BigData 2022 link(go to the website)