Učení umělých neuronových sítí založené na mikrogenetických a genetických algoritmech a ověření na vysokofrekvenčních finančních datech

Abstract

Predictions of currency exchange rates and stock values are very complex problems. Their solutions benefit from the state-of-the-art information technology and research and development of machine learning and artificial intelligence in general. The main focus of this study is to create and implement a learning algorithm for artificial neural networks which would effectively and efficiently train the network on large datasets. Trained networks are intended to be used as high-frequency economic data predictors. A modification of a micro-genetic algorithm for higher efficiency on contemporary computer hardware is created and described in this study. This modified micro-genetic algorithm is then compared to the original micro-genetic algorithm. The micro-genetic algorithm was more efficient while the predictions made by neural networks trained by both the modified micro-genetic and the original micro-genetic algorithms were equally good (equally precise). Since additional values can be added to the training dataset during training, these networks can be used for real-time predictions and decisions, e. g. for high-frequency trading