Skip to content

Kappa is a similar measure to accuracy(), but is normalized by the accuracy that would be expected by chance alone and is very useful when one or more classes have large frequency distributions.


kap(data, ...)

# S3 method for data.frame
  weighting = "none",
  na_rm = TRUE,
  case_weights = NULL,

  weighting = "none",
  na_rm = TRUE,
  case_weights = NULL,



Either a data.frame containing the columns specified by the truth and estimate arguments, or a table/matrix where the true class results should be in the columns of the table.


Not currently used.


The column identifier for the true class results (that is a factor). This should be an unquoted column name although this argument is passed by expression and supports quasiquotation (you can unquote column names). For _vec() functions, a factor vector.


The column identifier for the predicted class results (that is also factor). As with truth this can be specified different ways but the primary method is to use an unquoted variable name. For _vec() functions, a factor vector.


A weighting to apply when computing the scores. One of: "none", "linear", or "quadratic". Linear and quadratic weighting penalizes mis-predictions that are "far away" from the true value. Note that distance is judged based on the ordering of the levels in truth and estimate. It is recommended to provide ordered factors for truth and estimate to explicitly code the ordering, but this is not required.

In the binary case, all 3 weightings produce the same value, since it is only ever possible to be 1 unit away from the true value.


A logical value indicating whether NA values should be stripped before the computation proceeds.


The optional column identifier for case weights. This should be an unquoted column name that evaluates to a numeric column in data. For _vec() functions, a numeric vector.


A tibble with columns .metric, .estimator, and .estimate and 1 row of values. For grouped data frames, the number of rows returned will be the same as the number of groups. For kap_vec(), a single numeric value (or NA).


Kappa extends naturally to multiclass scenarios. Because of this, macro and micro averaging are not implemented.


Cohen, J. (1960). "A coefficient of agreement for nominal scales". Educational and Psychological Measurement. 20 (1): 37-46.

Cohen, J. (1968). "Weighted kappa: Nominal scale agreement provision for scaled disagreement or partial credit". Psychological Bulletin. 70 (4): 213-220.

See also

Other class metrics: accuracy(), bal_accuracy(), detection_prevalence(), f_meas(), j_index(), mcc(), npv(), ppv(), precision(), recall(), sens(), spec()


Max Kuhn

Jon Harmon



# Two class
kap(two_class_example, truth, predicted)
#> # A tibble: 1 × 3
#>   .metric .estimator .estimate
#>   <chr>   <chr>          <dbl>
#> 1 kap     binary         0.675

# Multiclass
# kap() has a natural multiclass extension
hpc_cv %>%
  filter(Resample == "Fold01") %>%
  kap(obs, pred)
#> # A tibble: 1 × 3
#>   .metric .estimator .estimate
#>   <chr>   <chr>          <dbl>
#> 1 kap     multiclass     0.533

# Groups are respected
hpc_cv %>%
  group_by(Resample) %>%
  kap(obs, pred)
#> # A tibble: 10 × 4
#>    Resample .metric .estimator .estimate
#>    <chr>    <chr>   <chr>          <dbl>
#>  1 Fold01   kap     multiclass     0.533
#>  2 Fold02   kap     multiclass     0.512
#>  3 Fold03   kap     multiclass     0.594
#>  4 Fold04   kap     multiclass     0.511
#>  5 Fold05   kap     multiclass     0.514
#>  6 Fold06   kap     multiclass     0.486
#>  7 Fold07   kap     multiclass     0.454
#>  8 Fold08   kap     multiclass     0.531
#>  9 Fold09   kap     multiclass     0.454
#> 10 Fold10   kap     multiclass     0.492