Institutional Repository of Vilnius University
Žmogui kenksmingų mikroorganizmų spektrų klasifikavimas mašinų mokymo metodais /
Abstract
dc:descriptionTraditional methods for microorganism identification, such as culture cultivation, are time-consuming and require specific laboratory conditions. Modern methods like MALDI-TOF MS and electrophoresis offer faster results but involve expensive equipment and complex sample handling. Fourier-Transform Infrared Spectroscopy Attenuated Total Reflectance (FTIR-ATR) is a promising alternative due to its sensitivity and speed, though it struggles with distinguishing closely related species. Machine learning, specifically supervised learning techniques, is explored to enhance the classification accuracy of microorganisms using spectral data. The methods discussed include Linear Regression, k-Nearest Neighbours, Decision Tree, AdaBoost, Gradient Boosting, Random Forests, and Neural Networks. The aim of this study is to evaluate the applicability of machine learning methods for improving the identification of harmful microorganisms in mixed samples by assessing their class distribution in the samples based on spectrographic data. The study involves preprocessing spectral data using Principal Component Analysis (PCA) to reduce dimensionality and highlight significant features. Multiple machine learning models are trained and evaluated, including Gradient Boosting, which iteratively reduces prediction error, and AdaBoost, which adapts to improve weaker models. The Gradient Boosting model demonstrated the best performance with the lowest mean square error, suggesting its effectiveness in predicting the correct mixture ratios of microorganism samples. AdaBoost showed discontinuities in its results, indicating a strong optimization towards specific value matches, making it seem more suitable for classification rather than regression tasks. The application of PCA improved the results of all models, particularly highlighting the significant spectral regions for interpretation. While the Gradient Boosting model performed best, the overall differences among the models were not substantial. Machine learning models achieved results with a root mean squared error of 0.125 to 0.166 and a mean absolute error of 0.097 to 0.114. This level of accuracy is sufficient to determine whether a significant amount of the second pathogen is present in the mixture and to estimate the ratio in which the two pathogens are distributed in the mixture. All models, except for the neural network, achieved similar results, so it cannot be stated that any are better suited to the task of estimating the ratio of harmful microorganisms in mixtures from spectra. The accuracy of the data used in the study had an impact on the achievable accuracy of the models. A repeat study with more accurate data would allow for a better assessment of the differences between models.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Motiejūnas, Benas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD210579745&prefLang=en_US