Back to search

Institutional Repository of Vilnius University

Ramano spektroskopijos duomenų klasifikavimas esant nesubalansuotoms klasėms /

Abstract

dc:description

Raman spectroscopy is often used in food industry for analyzing products and their quality. One of the most popular cases in food products analysis is to separate different types or classes of the same product to declare uses, quality or benefit for humans. But while Raman spectroscopy is expensive method to analyze products or some data is hard to find or measure there might be problem that in some cases there is not enough data to correctly classify different types of products. In this bachelor thesis the honey Raman spectra dataset is used where types of honey has different data quantity. The main goal is to find methods that can solve data shortage problem when there is no way to collect more data and as a consequence dataset has imbalanced classes. To achieve this goal three methods for handling imbalanced classes were selected: SMOTE, NearMiss and combination of both undersampling and oversampling methods. Then these three class balancing techniques were tested in supervised machine learning by using support-vector machine, logistic regression and decision tree classification algorithms. Firstly classification models were trained and tested with imbalanced dataset, then class balancing methods were applied to dataset and classifiers training ant testing were repeated. After all this process classification results before and after balancing classes were compared with each other. After experiments with selected class balancing methods and classification algorithms there was observed that SMOTE method in binary classification improved all models classification accuracy. Support-vector machine and decision tree classifiers accuracy was increased by even 9% and 12% accordingly and these models recognized minority class much better. And also it was noticed that for multiclass classification the best fitting method to balance the dataset is combination of oversampling and undersampling methods. By combining SMOTE and Random undersampling methods the accuracy score of support-vector machine and decision tree classifiers was increased up to 10%.

Degree

thesis:*
Grantor dc:publisher
Institutional Repository of Vilnius University
Year dc:date
2021

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Vencevičiūtė, Milda,
Contributors dc:contributor
  • Raila, Tomas

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

Vencevičiūtė, Milda,. Ramano spektroskopijos duomenų klasifikavimas esant nesubalansuotoms klasėms /. Institutional Repository of Vilnius University, 2021. https://repository.vu.lt/VU:ELABAETD107118422&prefLang=en_US