Razvoj algoritma za napovedovanje vrednosti in trgovanje s kriptovalutami

AŽMAN, BOŠTJAN (author), Hovelja, Tomaž (mentor)

Cilj diplomske naloge je razviti profitabilen algoritem za napovedovanje cen in trgovanje s kriptovalutami, osnovan na tehnologijah strojnega učenja in sentimentalne analize. V sklopu praktičnega dela raziskave bomo implementirali osnovne funkcionalnosti avtomatskega trgovanja, torej sposobnost programske opreme, da simulirano trguje z izbrano kriptovaluto. Funkcionalnost napovedovanja cen bomo nadgradili z logiko za trgovanje. Na problem napovedovanja cen lahko gledamo kot na problem napovedovanja časovnih vrst. Tovrstne probleme rešujemo z modeli, ki napovedujejo nove vrednosti na osnovi časovne vrste starih vrednosti. Reševanja tega problema se bomo lotili z uporabo spodbujevanega učenja in globokih q-nevronskih mrež. Izbrani problem je zahteven, saj produkt konkurira na odprtem trgu, kjer se izvajajo dovršene rešitve, v katere so podjetja vložila veliko sredstev. Iz tega razloga bomo za implementacijo uporabili programsko knjižnico z orodji za delo z modeli spodbujevanega učenja in s tem znižali kompleksnost implementacije. Programska oprema bo v celoti razvita v programskem jeziku python.

• Language: Slovenian
• Keywords: avtomatizirani trgovalni sistem, spodbujevano učenje, sentimentalna analiza, globoke q-nevronske mreže
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2023
• PID: 20.500.12556/RUL-155812
• Publication date in RUL: 19.04.2024

Secondary language

• Language: English
• Title: Development of a price prediction and trading algorithm for cryptocurrencies

Abstract

The aim of the diploma thesis is to develop an algorithm for accurate cryptocurrency price prediction and trading implemented using machine learning and sentimental analysis. In the practical part of the thesis we will implement an algorithm with core functionalities of automated trading meaning the model will have the ability to simulate trading of selected cryptocurrency. The functionality of price prediction has to be expanded with a trading strategy in order to trade successfully. The problem of price forecasting is best described by time series forecasting. Such problems can be solved with a model that predicts new values based on a time series of past values. We approached this problem with a model implemented with reinforcement learning, sentimental analysis and deep q-networks. Trading with profit is a complex task because our model competes with state of the art proprietary algorithms on the open market. For this reason we chose to implement our algorithm following modern reinforcement learning architecture with python’s several machine learning libraries.

• Keywords: automated trading system, reinforcement learning, sentimental anlysis, deep q-networks