Back to search

Institutional Repository of Vilnius University

Mašininio mokymosi modelių palyginimas skambučių srautui prognozuoti /

Abstract

dc:description

This bachelor's thesis examines the problem of call volume forecasting in financial institution contact centers by analyzing and comparing different models: the classical statistical SARIMAX model, and machine learning models: the tree-based XGBoost and the deep learning Gated Recurrent Unit (GRU) network. The aim of the study is to identify the most suitable forecasting approach that remains reliable under extreme operational conditions. The analysis is based on real data from a Lithuanian financial institution covering the period from 2020 to 2026, and model performance is evaluated using a rolling window approach during the extreme period of the 2nd Pillar Pension Reform (Q1 2026). The study applies formal testing methods: the Diebold–Mariano test with HLN correction to assess statistical significance, the Mann–Kendall trend test to analyze adaptation, and Winkler scores to evaluate the quality of prediction intervals. Model performance was assessed using a rolling window procedure with weekly coefficient updates, allowing for a more realistic evaluation of how models adapt to structural breaks. The results showed that under complex and rapidly changing operational conditions, a standalone XGBoost model, based on periodic lags and calendar variables, achieved the highest forecasting accuracy (MAPE 12.52%) and outperformed both the more complex GRU network and the hybrid (Stacking) model. However, the prediction interval coverage of XGBoost was unreliable, so for risk assessment it is safer to rely on intervals produced by the classical model. It was also observed that MinMaxScaler normalization performs worse when call volumes are asymmetric, indicating that standard transformations are not always suitable for deep learning models.

Degree

thesis:*
Grantor dc:publisher
Institutional Repository of Vilnius University
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Šimanskis, Gytis,

Rights

dc:rights
Statement dc:rights
  • info:eu-repo/semantics/openAccess
Language dc:language
lit

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
Vilnius University
Base URL
epublications.vu.lt/oai
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Šimanskis, Gytis,. Mašininio mokymosi modelių palyginimas skambučių srautui prognozuoti /. Institutional Repository of Vilnius University, 2026. https://repository.vu.lt/VU:ELABAETD308327887&prefLang=en_US