MSU Scientists Train Neural Network to Improve Financial Market Forecasts
Scientists at Moscow State University have combined a neural network with fuzzy logic methods, improving the accuracy of financial time-series forecasts by a factor of 1.3–1.7 compared with a baseline model.
The researchers from MSU’s Brain, Cognitive Systems, Artificial Intelligence interdisciplinary research and education school presented the development. Their paper was published in the Communications in Computer and Information Science series.
Conventional long short-term memory neural networks are sensitive to noise in data and use fixed activation functions. The researchers augmented the network with fuzzy logic, which accounts for data uncertainty both during preprocessing and within the network itself.
The method was tested on BTC–USD price data at one-minute, one-hour and one-day intervals. Forecast accuracy improved by factors of 1.3, 1.7 and 1.5, respectively.
