Skip to contents

The Brier score is a proper score function that measures the accuracy of probabilistic predictions. It is applicable to tasks in which predictions must assign probabilities to a set of mutually exclusive discrete outcomes. For binary classification, the Brier score is a measure of how far the predicted probabilities are from the actual outcomes.

Usage

dx_brier(predprob, truth, detail = "full")

Arguments

predprob

Numeric vector of predicted probabilities associated with the positive class.

truth

Numeric vector of true binary outcomes, typically 0 or 1, with the same length as predprob.

detail

Character specifying the level of detail in the output: "simple" for raw estimate, "full" for detailed estimate including 95% confidence intervals.

Value

Depending on the detail parameter, returns a numeric value representing the calculated metric or a data frame/tibble with detailed diagnostics including confidence intervals and possibly other metrics relevant to understanding the metric.

Details

The formula for the Brier score in a binary classification is:

$$BS = \frac{1}{N} \sum_{i=1}^{N} (f_i - o_i)^2$$

where:

  • \(N\) is the number of predictions,

  • \(f_i\) is the predicted probability of the occurrence of the positive class for the ith prediction,

  • \(o_i\) is the actual outcome for the ith prediction, 0 or 1.

The Brier score ranges from 0 to 1, where 0 represents a perfect model and 1 represents the worst model. It is equivalent to the mean squared error used in regression and can be decomposed into calibration loss, refinement loss, and uncertainty. This makes it a very informative metric for probabilistic forecasts, providing a nuanced view of the model's predictive performance.

Examples

predprob <- dx_heart_failure$predicted
truth <- dx_heart_failure$truth
simple <- dx_brier(predprob, truth, detail = "simple")
detailed <- dx_brier(predprob, truth)
print(simple)
#> [1] 0.1137428
print(detailed)
#> # A tibble: 1 × 8
#>   measure     summary estimate conf_low conf_high fraction conf_type notes      
#>   <chr>       <chr>      <dbl> <lgl>    <lgl>     <chr>    <chr>     <chr>      
#> 1 Brier Score 0.11       0.114 NA       NA        ""       ""        CIs not ye…