Metody strojového učení a umělé inteligence pro automatizované obchodování

Abstract

This bachelor's thesis focuses on the design and implementation of a universal automated cryptocurrency trading system utilizing artificial intelligence methods. The introductory section characterizes the specifics of the cryptocurrency market and the limitations of traditional analytical approaches. The theoretical overview describes various principles of artificial intelligence applications, presenting techniques based on natural language processing and time series prediction. The thesis also briefly reviews existing solutions. The practical part compares the predictive capabilities of ARIMA, MLP, and LSTM models, with the LSTM model achieving the best results. The final solution employs a microservices architecture leveraging Docker, Kubernetes, and Apache Kafka technologies. The conclusion discusses practical limitations of the implemented system related to transaction costs and outlines possibilities for future development.