magistrsko delo
Jana Janković (Author), Grega Vrbančič (Mentor)

Abstract

Živila so osnovne dobrine z velikim vplivom na gospodarsko in družbeno stabilnost, zato je natančno napovedovanje njihovih cen ključno. Modeli globokega učenja lahko prepoznajo kompleksne vzorce v časovnih vrstah, kot so zgodovinske cene živil. V tej raziskavi smo eksperimentalno primerjali konvencionalni pristop učenja in učenje s prenosom znanja v rekurentnih nevronskih mrežah za napovedovanje cen. Po iskanju optimalnih hiperparametrov smo modele naučili nad podatki, uporabili prenos znanja in ovrednotili oba pristopa. Na podlagi pridobljenih rezultatov smo ugotovili, da učenje s prenosom znanja bistveno pospeši proces učenja, vendar na račun slabše napovedne uspešnosti. Kljub temu pa rezultati magistrskega dela prispevajo k razumevanju, kdaj, zakaj in v kakšnih primerih je uporaba učenja s prenosom znanja smiselna izbira.

Keywords

napovedovanje cen živil;učenje s prenosom znanja;analiza časovnih vrst;magistrske naloge;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher: [J. Janković]
UDC: 004.85.032.26(043.2)
COBISS: 225867523 Link will open in a new window
Views: 0
Downloads: 12
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Utilization of transfer learning with recurrent neural networks for grocery price forecasting
Secondary abstract: Food is a fundamental commodity with a significant impact on economic and social stability, making accurate price forecasting essential. Deep learning models can identify complex patterns in time series, such as historical food prices. In this study, we experimentally compared the conventional learning approach with transfer learning in recurrent neural networks for price forecasting. After identifying optimal hyperparameters, we trained the models, applied transfer learning, and evaluated both approaches. Based on the obtained results, we found that transfer learning significantly accelerates the learning process, though at the cost of predictive performance. Nevertheless, the results of this master’s thesis contribute to understanding when, why, and in what scenarios transfer learning is a sensible choice.
Secondary keywords: food price prediction;transfer learning;RNN;time series analysis;
Type (COBISS): Master's thesis/paper
Thesis comment: Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Informatika in tehnologije komuniciranja
Pages: 1 spletni vir (1 datoteka PDF (XI, 73 str.))
ID: 25451064