muod function

Massive Unsupervised Outlier Detection (MUOD)

MUOD finds outliers by computing for each functional data, a magnitude, amplitude and shape index. Outliers are then detected in each set of index and outliers found are classified as either a magnitude, shape or amplitude outlier.

muod(dts, cut_method = c("boxplot", "tangent"))

Arguments

• dts: a matrix or dataframe of size n observation by p domain points.
• cut_method: a character value indicating method to use for finding indices cutoff. Must be either "boxplot" or "tangent".

Returns

Returns a list containing the following - outliers: a vector containing the indices of outliers identified.

• indices: a dataframe containing the shape, magnitude and amplitude indices

Details

MUOD was proposed in Azcorra et al. (2020) tools:::Rd_expr_doi("10.1038/s41598-018-24874-2")

as a support method for finding influential users in a social network data. It was also mentioned in Vinue and Epiphano (2020) tools:::Rd_expr_doi("10.1007/s11634-020-00412-9")

where it was compared with other functional outlier detection methods.

MUOD computes for each curve three indices: amplitude, magnitude and shape indices. Then a cutoff is determined for each set of indices and outliers are identified in each set of index. Outliers identified in the magnitude indices are flagged as magnitude outliers. The same holds true for the amplitude and shape indices. Thus, the outliers are not only identified but also classified.

Examples

dt1 <- simulation_model1()
md <- muod(dts = dt1$data)
str(md$outliers)
dim(md$indices)

References

Azcorra, A., Chiroque, L. F., Cuevas, R., Anta, A. F., Laniado, H., Lillo, R. E., Romo, J., & Sguera, C. (2018). Unsupervised scalable statistical method for identifying influential users in online social networks. Scientific reports, 8(1), 1-7.