We present definitions and properties of the fast massive unsupervised outlierdetection (FastMUOD) indices, used for outlier detection (OD) in functional data.FastMUOD detects outliers by computing, for each curve, an amplitude, magnitude, andshape index meant to target the corresponding types of outliers. Some methodsadapting FastMUOD to outlier detection in multivariate functional data are thenproposed. These include applying FastMUOD on the components of the multivariatedata and using random projections. Moreover, these techniques are tested on varioussimulated and real multivariate functional datasets. Compared with the state of the artin multivariate functional OD, the use of random projections showed the most effectiveresults with similar, and in some cases improved, OD performance. Based on the propor-tion of random projections that flag each multivariate function as an outlier, we proposea new graphical tool, the magnitude-shape-amplitude (MSA) plot, useful for visualizingthe magnitude, shape and amplitude outlyingness of multivariate functional data.
Classification
subjects
Computer Science
Statistics
keywords
fastmuod; functional data; functional outlier detection; multivariate functional data; outlierclassification; video data