confusion function

Given actual and predicted group assignments, give the confusion matrix

confusion(actual, predicted, gpnames = NULL, rowcol=c("actual", "predicted"),
printit = c("overall","confusion"), prior = NULL, digits=3)

Arguments

• actual: Actual (prior) group assigments
• predicted: Predicted group assigments.
• gpnames: Names for groups, if different from levels(actual)
• rowcol: For predicted categories to appear as rows, specify rowcol="predicted"
• printit: Character vector. Print "overall", or "confusion" matrix, or both.
• prior: Prior probabilities for groups, if different from the relative group frequencies
• digits: Number of decimal digits to display in printed output

Details

Predicted group assignments should be estimated from cross-validation or from bootstrap out-of-bag data. Better still, work with assignments for test data that are completely separate from the data used to dervive the model.

Returns

A list with elements overall (overall accuracy), confusion (confusion matrix) and prior (prior used for calculation of overall accuracy)

References

Maindonald and Braun: 'Data Analysis and Graphics Using R', 3rd edition 2010, Section 12.2.2

Author(s)

John H Maindonald

Examples

library(MASS)
library(DAAG)
cl <- lda(species ~ length+breadth, data=cuckoos, CV=TRUE)$class
confusion(cl, cuckoos$species)

## The function is currently defined as
function (actual, predicted, gpnames = NULL,
            rowcol = c("actual", "predicted"),
            printit = c("overall","confusion"),
            prior = NULL, digits = 3) 
{
  if (is.null(gpnames)) 
    gpnames <- levels(actual)
  if (is.logical(printit)){
    if(printit)printit <- c("overall","confusion")
    else printit <- ""
  }
  tab <- table(actual, predicted)
  acctab <- t(apply(tab, 1, function(x) x/sum(x)))
  dimnames(acctab) <- list(Actual = gpnames, `Predicted (cv)` = gpnames)
  if (is.null(prior)) {
    relnum <- table(actual)
    prior <- relnum/sum(relnum)
    acc <- sum(tab[row(tab) == col(tab)])/sum(tab)
  }
  else {
    acc <- sum(prior * diag(acctab))
  }
  names(prior) <- gpnames
  if ("overall"%in%printit) {
    cat("Overall accuracy =", round(acc, digits), "\n")
    if(is.null(prior)){
      cat("This assumes the following prior frequencies:", 
          "\n")
      print(round(prior, digits))
    }
  }
  if ("confusion"%in%printit) {
    cat("\nConfusion matrix", "\n")
    print(round(acctab, digits))
  }
  invisible(list(overall=acc, confusion=acctab, prior=prior))
}