Flag regression training outliers

Adds robust response, local-outlier-factor, and optional Cook's-distance scores to a regression training table.

Usage

detectOutliers(DT, threshold.mad = 3.5, threshold.lof = 2.0,
  threshold.cooks = Inf, minPts = 5, use.cooks = FALSE)

Arguments

DT

Training data.table containing numeric response column y.

threshold.mad

Modified Z-score threshold.

threshold.lof

Local outlier factor threshold.

threshold.cooks

Cook's distance threshold, or Inf for default rule.

minPts

Minimum points for local outlier factor.

use.cooks

Logical; include Cook's distance.

Value

A copy of DT with .score.mad, .score.lof, and .outlier columns; .score.cooks is added when requested.