@misc{Kerst, type = {Master Thesis}, author = {Kerst, Mario}, title = {Validation and optimization of mathematical predictive models for the identification of microorganisms using Raman-spectroscopy}, url = {http://nbn-resolving.de/urn:nbn:de:hbz:1383-opus4-14820}, school = {Hochschule Rhein-Waal}, pages = {86}, abstract = {The identification and classification of microorganisms remains a challenge in many areas, especially considering time and cost-efficiency. Fast, reliable and cheap methods are of great interest for research, industry and health care. Raman spectroscopy is a method that fulfills these criteria and, in this work, we aim to show its effectiveness to classify a selection of microorganisms. Using a standardized protocol, eighteen microorganisms were measured using Raman spectroscopy and classified with several types of mathematical classification models, including linear discriminant analysis, support vector machines and convolutional neural networks. To prepare the measured spectral data for classification several steps for the data evaluation and transformation were applied and their effectiveness for removing noise and other interferences evaluated. The resulting predictions outcomes were compared to select the most efficient model, showing that with an overall true positive prediction rate of 86.87 \% the convolutional neural network performed the best. Further inspection of the results shows issues with the measurements of selected organisms, indicating that an improved measurement protocol is required to achieve proper predictions. Overall this work shows that Raman spectroscopy in combination with mathematical classification models is a viable and fast method for classification of microorganisms.}, language = {en} }