Predviđanje cijena dionica korištenjem LSTM mreže

undergraduate thesis

• Title (english): Predicting Stock Prices Using the LSTM Recurrent Neural Network
• Author: Filip Radović
• Mentor: Klemo Vladimir (mentor)
• Committee member: Klemo Vladimir (predsjednik povjerenstva)
• Committee member: Adrian Satja Kurdija (član povjerenstva)
• Committee member: Domagoj Alagić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-17, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Tema ovog rada je izrada LSTM modela za predviđanje cijena dionica. Na početku je dan kratki uvod u povijest ovog problema i u rješenja koja su do sada predložena. Opisani su klasični problemi s kojima se susrećemo pri izradi LSTM modela za predviđanje cijena dionica. Pokazan je način na koji se može izbjeći problem autokorelacije vremenskih serija koji uzrokuje netočne rezultate kod treniranja LSTM mreža. Opisana je tehnika kojom se
mogu normalizirati podaci dionica. Prikazani su alternativni izvori podataka za predviđanje cijena dionica. Detaljno su opisane tehnike za uklanjanje šuma iz podataka poput wavelet transformacije i autoenkodera. Opisane su arhitekture korištenih mreža i algoritmi korišteni za njihovu optimizaciju. Testirane su različite kombinacije hiperparametara i objašnjeni rezultati koji su dobiveni pomoću njih.

Abstract (english)

The subject of this paper is the creation of an LSTM model for forecasting stock prices. The paper gives a brief introduction to the history of the problem and describes solutions that have been suggested so far. Classical problems that appear when forecasting stock prices with an LSTM model are described. A way to eliminate the problem arising from the auto-correlation of time series data is demonstrated. A successful normalization technique for stock data is described. Alternative data sources for forecasting stock prices are shown. Comprehensive descriptions of noise removal techniques from data such as wavelet transformation and autoencoders are given. The architecture of the networks and the algorithms used for their optimization are described. Various combinations of hyperparameters are explored, and the results obtained by using them are given.

• Keywords: umjetna inteligencija; neuronske mreže; autoenkoder; LSTM mreža; predviđanje cijena dionica; wavelet transformacija; uklanjanje šuma iz signala; vremenska serija
• Keywords (english): artificial intelligence; neural networks; autoencoder; LSTM network; stock price forecasting; wavelet transformation; signal noise removal; time series
• Language: croatian
• URN:NBN: urn:nbn:hr:168:725275
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-01-07 21:44:22