Skip to contents

Combined outlier detection using multiple methods

Source: R/mt_combined_outliers.R
mt_combined_outliers.Rd

Runs the simultaneous method (with gap and entropy thresholds) and the sequential scan, then combines their votes. A location flagged by at least min_votes methods is declared an outlier.

Usage

mt_combined_outliers(
  x,
  reference = NULL,
  min_votes = 2,
  scan = "forward-backward",
  time_normalize = TRUE,
  plot = FALSE,
  ...
)

Arguments

x

A move2 object.

reference

Optional move2 object for reference distributions.

min_votes

Minimum number of methods (out of 3) that must flag a location for it to be declared an outlier. Default: 2 (majority vote).

scan

Scanning strategy for the sequential method: "forward-backward", "greedy", or "random".

time_normalize

Logical; if TRUE (default), use speed and angular velocity.

plot

Logical; if TRUE, plot the results.

...

Additional arguments passed to mt_flag_outliers and mt_sequential_outliers.

Value

The input move2 object with added columns: is_outlier, vote_count (0–3), and the individual method flags flag_gap, flag_entropy, flag_seq.

Details

Relationship to the four-primitive cascade. This function votes across three strategies on a single detector (the joint-probability surface from mt_flag_outliers): simultaneous-with-gap, simultaneous-with-entropy, and the sequential scan from mt_sequential_outliers. The unified cleaner mt_clean_track is a different construction: it votes across four detectors (bridge, detour, probability, speed-cap), each looking at a different grain of the data. The two are not redundant – the cascade cannot reach inside a single detector to compare simultaneous vs sequential evaluation. Reach for mt_combined_outliers when you want to inspect the per-strategy agreement on the probability surface specifically, or when calibrating thresholds on that surface and wanting cross-strategy validation. For routine cleaning, mt_clean_track is the recommended entry-point.

References

Safi, K. (in preparation). Self-thresholding hierarchical outlier-detection for animal movement tracks. Companion paper to the move2utils R package. Preprint: bioRxiv (DOI forthcoming).

See also

mt_clean_track (recommended unified cleaner; votes across four detectors rather than three strategies on one detector); mt_flag_outliers (the probability primitive this function votes strategies on); mt_sequential_outliers (one of the three voted strategies); mt_persistence_score (multi-scale persistence annotation that can be applied to this function's output for additional confidence quantification).

Examples

if (FALSE) { # \dontrun{
## Majority-vote flagging across three detection strategies:
res <- mt_combined_outliers(track, min_votes = 2,
                             scan = "forward-backward")
table(res$vote_count)
} # }