Comparison of Outlier Detection Methods on Astronomical Image Data

Among the many challenges posed by huge data volumes produced new generation of astronomical instruments there is also search for rare and peculiar objects. Unsupervised outlier detection algorithms may provide a viable solution. In this work we compare performances six methods: Local Outlier Factor, Isolation Forest, k-means clustering, measure novelty, both normal convolutional autoencoder. These methods were applied to extracted from SDSS stripe 82. After discussing sensitivity each method its own set hyperparameters, combine results rank objects produce final list outliers.