Automated trading system using machine learning, stream mining and technical analysis of trading

Digitalno trgovanje z dobrinami (delnicami, valutami itd.) danes močno izpodriva klasično trgovanje, saj se je veliko borz preselilo v oblak. Računalnik poskrbi za izmenjavo dobrin, hkrati pa lahko tudi samostojno trguje. Avtomatski trgovalni sistem je računalniški sistem, ki trguje z dobrinami brez posredovanja človeka. Tak pristop poleg objektivnosti in empiričnosti odločitev omogoča hitre izvršitve kupčij, ki so ključne za uspeh. V diplomski nalogi raziskujemo avtomatske trgovalne sisteme in njihov način delovanja. Naredimo pregled področja tehnične analize trgovanja, ki se ukvarja s kvantificiranjem gibanja cen na trgu. Definiramo problem trgovanja z vidika nadzorovanega strojnega učenja in rudarjenja podatkovnih tokov. S področja strojnega učenja posebej izpostavimo algoritem k-najbližjih sosedov, umetne nevronske mreže in naivni Bayesov klasifikator. S pomočjo omenjenih algoritmov in podatkov, pridobljenih z metodami tehnične analize trgovanja, zasnujemo avtomatski trgovalni sistem. S simulacijo ga ovrednotimo na realnem gibanju cen kriptovalute Bitcoin in kriptovalute Litecoin. Avtomatsko trgovanje se z vidika strojnega učenja izkaže za zelo zahteven problem. Kljub temu nam uspe, z uporabo algoritma k-najbližjih sosedov in umetnih nevronskih mrež, doseči zadovoljivo napovedno uspešnost in posledično profitabilnost.Digital trading of securities is beginning to dominate over classical trading and the trading exchanges are rapidly migrating to the cloud. Computer is not only present in the exchange process but is also capable of making trading decisions on human\u27s behalf. Automated trading system is a computer system, capable of trading without human interaction. The benefits of such an approach to trading are objectivity and fast execution of orders, which are often crucial for success. In this thesis we examine automated trading systems and their structure. We study the field of technical analysis which quantifies market price movements. We define trading as a supervised machine learning and stream mining problem and examine the k-nearest neighbours algorithm, naive Bayes classifier and artificial neural networks. Based on our research we design an automated trading system. We evaluate its performance on actual market data of cryptocurrencies Bitcoin and Litecoin using a simulated environment. Automated trading turns out to be a difficult machine learning problem, but with the use of the k-nearest neighbours algorithm and artificial neural networks we manage to achieve decent success in our predictions and profitability