Outlier mining is an important data analysis task to distinguish exceptional outliers from regular objects. For outlier mining in the full data space, there are well established methods which are successful in measuring the degree of deviation for out lier ranking. However, in recent applications traditional outlier mining approaches miss outliers as they are hidden in subspace projections. Especially, outlier ranking approaches measuring deviation on all available attributes miss outliers deviating from their local neighborhood only in subsets of the attributes. In this work, we propose a novel outlier ranking based on the objects deviation in a statistically selected set of relevant subspace projections. This ensures to find objects deviating in multiple relevant subspaces, while it excludes irrelevant projec tions showing no clear contrast between outliers and the residual objects. Thus, we tackle the general challenges of detecting outliers hidden in subspaces of the data. We provide a selection of subspaces with high contrast and propose a novel ranking based on an adaptive degree of deviation in arbitrary subspaces.In thorough experiments on real and synthetic data we show that our approach outperforms competing outlier ranking approaches by detecting outliers in arbitrary subspace projections.

No abstract

Outlier mining is an important data analysis task to distinguish exceptional outliers from regular objects. In recent research novel outlier ranking methods propose to focus on outliers hidden in subspace projections of the data. However, focusing only on the detection of outliers these approaches miss to provide reasons why an object should be considered as an outlier. In this work, we propose a novel toolkit for exploration of subspace outlier rankings. To enable exploration of subspace outliers and to complete knowledge extraction we provide further descriptive information in addition to the pure detection of outliers. As wittinesses for the outlierness of an object, we provide information about the relevant projections describing the reasons for outlier properties. We provided SOREX as open source framework on our website 1 it is easily extensible and suitable for research and educational purposes in this emerging research area.

Different endotypes of asthma were described in human. Atopic asthma is a T-helper 2 (Th2)-mediated disease consisting mainly of an eosinophilic inflammation in the airways. Other endotypes show neutrophilic inflammation of the airways that is probably based on a Th17 response. There are several mouse models described in the literature to study the Th2 polarized eosinophilic disease, however, only a few models are available which characterize the neutrophilic endotype. The aim of this study was to compare both endotypes in relation to the severity of the allergen-induced inflammation. Groups of either Balb/c or DO11.10 mice were sensitized with ovalbumin (OVA) adsorbed to aluminum hydroxide. Mice were subsequently challenged with OVA for different periods of time. They were evaluated for airway hyperreactivity (AHR), cytokine production, airway inflammation, and remodeling of the airways. As expected, Balb/c mice developed a Th2 response with AHR, eosinophilic airway inflammation, and allergen-specific IgE and IgG1. By contrast DO11.10 mice showed a mixed Th1/Th17 response with strong neutrophilic airway inflammation, IgG2a, but only limited induction of AHR. While Balb/c mice showed remodeling of the airways with subepithelial fibrosis and goblet cell metaplasia, airway remodeling in DO11.10 mice was marginal. Both airway inflammation and remodeling resolved after prolonged periods of challenge in both models. In conclusion, strong allergen-induced airway remodeling in mice seems to be triggered by the specific conditions arising from infiltration with eosinophilic granulocytes in the lung. A Th1/Th17 response leading to neutrophilic inflammation does not seem to be sufficient to induce pronounced airway remodeling.