outlierspinner - Geometric Multivariate Outlier Detection via Random Directional
Probing
Provides tools for multivariate outlier detection based on
geometric properties of multivariate data using random
directional projections. Observation-level outlier scores are
computed by jointly probing radial magnitude and angular
alignment through repeated projections onto random directions,
with optional robust centering and covariance adjustment. In
addition to global outlier scoring, the method produces
dimension-level contribution measures to support interpretation
of detected anomalies. Visualization utilities are included to
summarize directional contributions for extreme observations.
