Skip to contents

Classification accuracy and agreement estimates

class_acc.Rd

Classification accuracy and agreement estimates for 2x2 contingency tables.

Usage

class_acc(x, conf.level = 0.95)

Arguments

x

A table in the correct format. See details.

conf.level

The confidence level as a proportion.

Value

data.frame

Details

The table must be in the following format for correct results:

actual +actual -
estimate +ab
estimate -cd

Wilson confidence intervals are returned for the proportion estimates. Functions from the DescTools package are used for confidence intervals. Estimates returned include:

  • Estimated prevalence: The estimated proportion of subjects with positive outcome (based on the estimated label).

  • Actual prevalence: The actual proportion of subjects with positive outcome (based on the actual label).

  • Estimated accuracy: Among all subjects, the proportion who had a correct classification.

  • Sensitivity: Among subjects with actual positive outcome, the proportion that were correctly classified with positive outcome.

  • Specificity: Among subjects with actual negative outcome, the proportion that were correctly classified with negative outcome.

  • Positive predictive value: Among subjects with estimated positive outcome, the proportion that were correctly classified with positive outcome. (Accuracy of a positive test)

  • Negative predictive value: Among subjects with estimated negative outcome, the proportion that were correctly classified with negative outcome. (Accuracy of a negative test)

  • Proportion of false positives: Among subjects with actual negative outcome, the proportion that were incorrectly classified as positive outcome.

  • Proportion of false negative: Among subjects with actual positive outcome, the proportion that were incorrectly classified as negative outcome.

  • Cohen's Kappa: A measure of agreement between the two tests that adjusts for the amount of agreement that would be expected due to chance alone.

Examples


#----------------------------------------------------------------------------
# class_acc() examples
#----------------------------------------------------------------------------
library(bkstat)

# simulate random data
new_test <- sample(
  x = c("positive", "negative"),
  size = 200,
  replace = TRUE,
  prob = c(0.25, 0.75)
)
standard_test <- sample(
  x = c("positive", "negative"),
  size = 200,
  replace = TRUE,
  prob = c(0.25, 0.75)
)

# ensure correct matrix structure order
new_test <- factor(new_test, levels = c("positive", "negative"))
standard_test <- factor(standard_test, levels = c("positive", "negative"))

tab <- table(new_test = new_test, standard_test = standard_test)

# results
class_acc(x = tab)
#>                          summary   estimate    lower_ci  upper_ci
#> 1            Apparent prevalance 0.27500000  0.21779700 0.3406834
#> 2                True prevalance 0.27000000  0.21323450 0.3354344
#> 3            Diagnostic accuracy 0.63500000  0.56631703 0.6985947
#> 4                    Sensitivity 0.33333333  0.22241417 0.4663904
#> 5                    Specificity 0.74657534  0.67032209 0.8101858
#> 6      Positive predictive value 0.32727273  0.21813066 0.4589678
#> 7      Negative predictive value 0.75172414  0.67552026 0.8149345
#> 8  Proportion of false positives 0.25342466  0.18981423 0.3296779
#> 9  Proportion of false negatives 0.66666667  0.53360964 0.7775858
#> 10                 Cohen's kappa 0.07944515 -0.06370327 0.2225936