This M.Sc. thesis report investigates the application of one-class classification techniques to complex high-dimensional data. The aim of a one-class classifier is to separate target data from non-target data, but only a dataset containing target data is available for training. The issue with high-dimensional data is that it is difficult to perform density estimation due to the `curse of dimensionality'. Most conventional method for one-class classification rely on density estimation. This thesis focusses on the use of autoencoders and generative adversarial networks (GANs) for one-class classification problems involving image data. Autoencoders can learn encoding and decoding functions for samples from the target dataset. These encoding and decoding functions are, however, expected to not perform well for non-target samples, as they have never been seen during the training phase. This makes it possible to separate target and non-target data. For GANs, the discriminator is used to distinguish between target and non-target data. Autoencoders and GANs are evaluated extensively in this report. Their behavior, desired parameters and strengths and weaknesses are evaluated by performing experiments. The main findings are that GANs do not perform well for one-class classification tasks, because of mode collapse and insufficient sampling of the non-target data. Even for extremely simple datasets these issues were observed. Autoencoders are shown to perform much better and behave according to the theoretical expectations.

Note: some parts of the abstract have been censored to hide company secrets. This project was performed for Orchid N.V., an innovator in agriculture. The project focuses on their Scalar system. This is a system that consists out of multiple machines that automatically feed cows. While farmers are generally pleased with these systems, ---- -- --- ---- - --- -- ------- -- --- ----------- -- --- ------. --- ------ ---- ------- - --- -- --- ----- --- ----- -- -- -------- ---- ----- ---- ---- -- -- -------------- --- -- --- ------. ---- ------- ----- -- ----- ---- ----- -- ---------- - -------- -------- ----- ------- ------- -- ---- ------- ---- ----- -------- ---- ---- ---- --- ------ -------- --- -------- --- ----------. A daily report about the performance of the systems on the farm was the initial assignment for the project. The project started with a two week research phase in which the assignment was further investigated by visiting testfarms and conducting interviews with farmers and Orchid employees. --- ----- --------- --- -- -- ---- -- ---------- ------ -- ------- --- ------- -- -------- -- ---- ------- ------- ---- ----- -------. ----- ---- --- --------- farmers had to be informed about potential problems on their farm. This can be done by looking at patterns and outliers. ------ ---- ----- - -------- --- ---- ---- -- ------- -- ---.--- --- ------------ ---- ---- -- ----- ----- -- ---------- --- -------- -------- --------- -- ----. After the research phase, three development cycles were executed. At the end of each design cycle, a new version of the product was reviewed within Orchid and at a testfarm. The final product consists out of a cloud solution that processes logs to generate data and a web application dashboard, which visualizes this data. The web application dashboard shows an overview of the performance of the Scalar system on the farm and a list of points of attention, with interesting findings within the data. These could be potential problems. The reactions from Orchid and the farm were very positive. During the process there were not many problems and the project can be considered a success. -------------- --- --- ---- -- --- ----- -- --------- --- ------------ --------- --- ---- --- --------- ----------- --- --- -------- --------- ------- ----------- --- ------------ -- --------- --- -- --------. The project was finalized with a final presentation at Orchid and a final presentation at TUDelft.