Usporedba algoritama za predviđanje cijena kriptovaluta

master's thesis

• Title (english): Comparison of cryptocurrency price prediction algorithms
• Author: Goran Brkić
• Mentor: Martina Antonić (mentor)
• Committee member: Martina Antonić (predsjednik povjerenstva)
• Committee member: Krešimir Pripužić (član povjerenstva)
• Committee member: Dragan Jevtić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-11, Croatia

Abstract

Ovaj rad istražuje primjenu različitih modela strojnog učenja za predviđanje smjera kretanja cijene kriptovaluta. Analizirani su modeli logističke regresije, SVM-a, slučajne šume, XGBoosta i LSTM-a koristeći preciznost, odziv i F1-ocjenu kao mjeru uspješnosti. Analiza je provedena nad kriptovalutnim parovima BTC/USDT, ETH/USDT i LINK/USDT kako bi se dodatno saznalo je li predviđanje kriptovaluta niže tržišne kapitalizacije jednostavnije ili teže. Kroz pregled relevantne literature, tehničke analize cijene, implementaciju pet različitih modela strojnog učenja te analizu dobivenih rezultata pruža se pregled mogućnosti korištenja strojnog učenja za predviđanje kretanja kriptovaluta. Rezultati pokazuju da navedeni modeli pružaju ograničenu prednost u odnosu na nasumično predviđanje smjera te podržavaju teoriju slučajnog hoda. Također, rezultati pokazuju smanjenu sposobnost predviđanja kriptovalutnog para manje tržišne kapitalizacije. Istraživanje pruža korisne smjernice za implementaciju navedenih modela te naglašava potrebu za daljnjim optimiziranjem modela koji je postigao najveću točnost, XGBoost.

Abstract (english)

This paper explores different machine learning models and their ability to predict the direction of price movement for different cryptocurrencies. Analysis was conducted on models of logistic regression, SVM, random forest, XGBoost and LSTM using precision, recall and F1-score as a method of comparison. Analysis was done on cryptocurrency pairs BTC/USDT, ETH/USDT and LINK/USDT to find out whether cryptocurrencies with lower market capitalization are harder to predict. Through a review of relevant literature, technical price analysis, the implementation of five different machine learning models, and the analysis of the obtained results, this paper provides an overview of the possibilities of using machine learning for predicting cryptocurrency movements. The results indicate that these models offer a limited advantage compared to random prediction and support the random walk theory. Additionally, the findings suggest a reduced ability to predict cryptocurrencies with lower market capitalizations. This research offers useful guidelines for implementing the mentioned models and emphasizes the need for further optimization of the model that achieved the highest accuracy, XGBoost.

• Keywords: strojno učenje; kriptovalute; logistička regresija; SVM; slučajna šuma; XGBoost; LSTM; predviđanje cijene; burza; tehnički indikatori
• Keywords (english): machine learning; cryptocurrency; logistic regression; SVM; random forest; XGBoost; LSTM; price prediction; exchange; technical indicators
• Language: croatian
• URN:NBN: urn:nbn:hr:168:997389
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2024-10-12 20:12:05