Predikcija cijena vrijednosnica zasnovana na Dynamic Time Warping udaljenosti

undergraduate thesis

• Title (english): Prediction of Asset Prices Based on Dynamic Time Warping Distance
• Author: Lovro Mutvar
• Mentor: Zvonko Kostanjčar (mentor)
• Committee member: Zvonko Kostanjčar (predsjednik povjerenstva)
• Committee member: Lovre Mrčela (član povjerenstva)
• Committee member: Bruno Gašperov (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-02, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Cilj ovog rada bila je predikcija smjera kretanja cijena dionica na temelju Dynamic Time
Warping udaljenosti vremenskih nizova. Ulazni podaci bili su cijene 368 dionica iz S&P
(Standard & Poor's) indeksa u razdoblju od 3.6.2013. do 6.4.2017., ukupno 970 radnih
dana. U okviru ovog rada implementirani su i detaljno objašnjeni klasifikacijski algoritmi
strojnog učenja: algoritam k – najbližih susjeda (engl. k – nearest neighbours, KNN) i
algoritam stroja potpornih vektora (engl. Support vector machine, SVM). Navedeni
algoritmi klasificiraju vremenske nizove na temelju njihove udaljenosti dobivene pomoću
DTW algoritma. Klasifikacijski modeli trenirani su i testirani na ulaznim podacima te su
primijenjene različite metrike za evaluaciju njihove uspješnosti. Kao i što je očekivano
algoritam SVM sa univarijatim nizovima dao je bolje rezultate od algoritma KNN. Unatoč
očekivanjima, SVM sa multivarijatnim nizovima, koji koristi više informacija te u obzir
uzima cijene svih dionica u određenom trenutku, pokazao se kao lošiji model za predikciju
od SVM-a sa univarijatnim nizovima.

Abstract (english)

The aim of this paper was to predict the direction of stock price movement based on
Dynamic Time Warping distance of time series. The input dataset were prices of 368
shares from the S&P (Standard & Poor's) index in the period from 3.6.2013. to 6.4.2017,
which is a total of 970 working days. Within this paper, two machine learning
classification algorithms have been implemented and explained in detail: the k - nearest
neighbor algorithm (KNN) and the Support vector machine (SVM) algorithm. These
algorithms classify time series based on their distance obtained using the DTW algorithm.
Classification models were trained and tested on input data and various metrics were
applied to assess their performance. As expected, the SVM algorithm with univariate time
series gave better results than the KNN algorithm. Despite expectations, an SVM with
multivariate time series, which uses more information and takes into account all stock
prices at a given time, has proven to be a poorer model for predicting price movements
than the SVM algorithm with univariate time series.

• Keywords: vremenski niz; Dynamic Time Warping; klasifikacija; algoritam k – najbližih susjeda; algoritam stroj potpornih vektora
• Keywords (english): time series; Dynamic Time Warping; classification; k – nearest neighbours algorithm; Support vector machine algorithm
• Language: croatian
• URN:NBN: urn:nbn:hr:168:223265
• 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: 2023-01-14 21:16:44