Mean Absolute Percentage Error

Computes the mean absolute percentage error (MAPE) between predicted and reference values.

Usage

mape(x, ...)

# S3 method for class 'lm'
mape(x, ...)

# Default S3 method
mape(x, ref, na.rm = FALSE, ...)

Arguments

x

an object. Methods are available for numeric vectors and model objects (e.g. lm).

...

additional arguments passed to methods

ref

numeric vector of reference (true) values

na.rm

logical; whether to remove missing values

Value

a numeric scalar containing the MAPE, as a fraction, not a percentage: a mean absolute relative error of six percent is returned as 0.06. Multiply by 100 for the percentage form. The name is conventional; the definition below is the one implemented.

Details

The MAPE is defined as: $$ \frac{1}{n} \sum \left| \frac{ref - x}{ref} \right| $$

Note that values where ref = 0 lead to division by zero and result in NA.

See also

Other model.metrics: auc(), averagePrecision(), brierScore(), logLoss(), mae(), mse(), nmae(), nmse(), rmse(), smape()

Examples

x <- c(2.5, 3.0, 2.8)
ref <- c(3.0, 2.5, 3.0)

mape(x, ref)
#> [1] 0.1444444

# with linear model
fit <- lm(mpg ~ hp, data = mtcars)
mape(fit)
#> [1] 0.1566944