Morphological identification and classification of anaerobic gut fungi with artificial intelligence algorithms

In this thesis, artificial intelligence algorithms were used primarily in the classification and morphological identification of anaerobic gut fungi. Artificial intelligence included the use of machine learning (ML) and deep learning (DL) algorithms in the classification of microorganisms. These algorithms were used to analyze the characteristics of microorganisms (genetic, morphological, biochemical, etc.) and to make accurate classifications. In particular, the analysis of genome data (DNA, RNA sequences) was used as an important method to determine the species of microorganisms. Convolutional Neural Networks (CNN) architectures; AlexNet, ResNet and GoogLeNet have significant potential in the field of deep learning, especially in the field of image recognition and classification. In this study, a data set consisting of a total of 3,934 images was given, including 1,129 images from the anaerobic gut fungus species called Caecomyces, 1,253 images from the anaerobic gut fungus species called Neocallimastix, 1,126 images from the anaerobic gut fungus species called Orpinomyces, and 426 images from the anaerobic gut fungus species called Promyces. These data sets were given to the important artificial intelligence architectures Alexnet, Googlenet and Resnet-101 and 3 feature matrices were obtained for each of them. The feature matrix obtained from the Alexnet architecture was run separately in all classification algorithms in the classification layer in the Matlab program we used and the results were compared and 6 ranking algorithms that made the highest rate of classification were determined. These ranking algorithms are Quadratic SVM, Fine KNN, Wide Neural Network, Cubic SVM, Supspace KNN and SVM Kernel algorithms. 3 different feature matrices taken from 3 different architectures were studied with 6 different ranking algorithms and a total of 18 results were obtained. Especially the 94% accuracy rate obtained with the Resnet-101 architecture and Cubic SVM algorithm was a very high rate and study result.