Linear and Quadratic Discriminant Analysis

Fits LDA or QDA (MASS::lda / MASS::qda) on selected numeric columns, reports training predictions with a confusion matrix and accuracy, and can classify new observations.

Usage

mv_discrim(
  data,
  group,
  cols,
  method = c("lda", "qda"),
  prior = NULL,
  newdata = NULL
)

Arguments

data

A data frame.

group

<tidy-select> The single grouping column (coerced to a factor; must not contain NA).

cols

<tidy-select> Numeric predictor columns; must not include the grouping column (exclude it with -group).

method

Either "lda" (default) or "qda".

prior

Optional prior probabilities for the groups, passed to MASS::lda() / MASS::qda().

newdata

Optional data frame with the same predictor columns to classify; see new_predictions in the return value.

Value

An object of class mv_discrim: a list with components predictions (tibble with .row, actual, predicted, and post_<level> posterior columns), confusion (long tibble with actual, predicted, n), scores (tibble with .row and LD1... — LDA only, NULL for QDA), structure (tibble with variable and LD1... for LDA, or per-group means for QDA), new_predictions (tibble with .row — the row number within newdata — plus predicted and posterior columns; rows of newdata with missing predictor values are skipped), and meta.

Examples

r = mv_discrim(iris, Species, Sepal.Length:Petal.Width)
r$meta$accuracy
#> [1] 0.98
r$confusion
#> # A tibble: 9 × 3
#>   actual     predicted      n
#>   <fct>      <fct>      <int>
#> 1 setosa     setosa        50
#> 2 versicolor setosa         0
#> 3 virginica  setosa         0
#> 4 setosa     versicolor     0
#> 5 versicolor versicolor    48
#> 6 virginica  versicolor     1
#> 7 setosa     virginica      0
#> 8 versicolor virginica      2
#> 9 virginica  virginica     49