fdars (version 0.3.3)

outliers.lrt: LRT-based Outlier Detection for Functional Data

Description

Detects outliers using the Likelihood Ratio Test approach based on Febrero-Bande et al. Uses bootstrap to estimate a threshold and iteratively removes curves exceeding this threshold. Implemented in Rust for high performance with parallelized bootstrap.

Usage

outliers.lrt(
  fdataobj,
  nb = 200,
  smo = 0.05,
  trim = 0.1,
  seed = NULL,
  percentile = 0.99
)

Value

A list of class 'outliers.fdata' with components:

outliers

Indices of detected outliers

distances

Normalized distances for all curves

threshold

Bootstrap threshold used

percentile

Percentile used for threshold

fdataobj

Original fdata object

Arguments

fdataobj

An object of class 'fdata'.

nb

Number of bootstrap replications for threshold estimation (default 200).

smo

Smoothing parameter for bootstrap noise (default 0.05).

trim

Proportion of curves to trim for robust estimation (default 0.1).

seed

Random seed for reproducibility.

percentile

Percentile of bootstrap distribution to use as threshold (default 0.99, meaning 99th percentile). Lower values make detection more sensitive (detect more outliers).

Examples

t <- seq(0, 1, length.out = 50)
X <- matrix(0, 30, 50)
for (i in 1:30) X[i, ] <- sin(2*pi*t) + rnorm(50, sd = 0.1)
# Add an outlier
X[1, ] <- X[1, ] + 3
fd <- fdata(X, argvals = t)
out <- outliers.lrt(fd, nb = 100)

# More sensitive detection
out_sensitive <- outliers.lrt(fd, nb = 100, percentile = 0.95)