Machine-Learning Software for Automated Trading With Cryptocurrencies

Cilj ovog rada bio je izrada sustava za automatizirano trgovanje kriptovalutama (točnije Bitcoinom) utemeljen na strojnom učenju koji se nalazi u Google Cloudu. Nakon prikupljanja potrebnih podataka i modeliranja sustava, on je trebao prvo saznati trenutno stanje tržišta. Kada to sazna, treba te podatke unijeti u istrenirani model, odnosno neuronsku mrežu, koja daje svoje predviđanje. Ako neuronska mreža i društveni pokazatelji daju pozitivni signal, sustav treba signalizirati walletu da postavi narudžbu za prethodno određen iznos Bitcoina. Vrsta narudžbe je takva da sustav ne mora brinuti u kojem trenutku treba prodati kupljene kriptovalute, nego samo traži pravi trenutak kada ih kupiti.Point of this thesis was building a machine learning software for algorithmic tra- ding with cryptocurrencies (more precisely Bitcoin) placed on Google Cloud. After gathering needed data and building the software, firstly it needed to get the data of the current market position. When it is done, that data is fed to previously trained ne- ural network which then predicts if the cryptocurrency should be bought. If the neural network and the social indicators give positive signal, the software sends a (trailing stop loss) order with predefined amount to the wallet. Order is set up that way so the software doesn’t need to worry about perfect time to sell the asset