mae {ie2misc}

Mean-absolute error (MAE)

Description

This function computes the mean-absolute error (MAE).

Usage

mae(predicted, observed, na.rm = FALSE)

Arguments

predicted

numeric vector that contains the model predicted data points (1st parameter)

observed

numeric vector that contains the observed data points (2nd parameter)

na.rm

logical vector that determines whether the missing values should be removed or not.

Details

(MAE) is expressed as

n^{-1} \sum \limits_{i=1}^n{ \left| P_i - O_i \right|}

n

the number of observations

P

the "model estimates or predictions"

O

the "thought-to-be reliable and pairwise matched observations"

MAE is fully discussed in the Willmott reference, including a comparison to root mean square error (RMSE).

Value

mean-absolute error (MAE) as a numeric vector using the same units as the given variables. The default choice is that any NA values will be kept (na.rm = FALSE). This can be changed by specifying na.rm = TRUE, such as mae(pre, obs, na.rm = TRUE).

References

Cort J. Willmott and Kenji Matsuura, "Advantages of the mean-absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance", Climate Research, Vol. 30: 79-82, 2005, https://web.archive.org/web/20230119121852/climate.geog.udel.edu/~climate/publication_html/Pdf/WM_CR_05.pdf. Retrieved thanks to the Internet Archive: Wayback Machine

See Also

mape for mean absolute percent error (MAPE), madstat for mean-absolute deviation (MAD), dr for "index of agreement (dr)", vnse for Nash-Sutcliffe model efficiency (NSE), and rmse for root mean square error (RMSE).

Examples

library("ie2misc")

obs <- 1:10 # observed
pre <- 2:11 # predicted
mae(pre, obs)


library("rando")

set_n(100) # makes the example reproducible
obs1 <- r_norm(.seed = 103) # observed
pre1 <- r_norm(.seed = 102) # predicted


# using the vectors pre1 and obs1
mae(pre1, obs1)


# using a matrix of the numeric vectors pre1 and obs1
mat1 <- matrix(data = c(obs1, pre1), nrow = length(pre1), ncol = 2,
   byrow = FALSE, dimnames = list(c(rep("", length(pre1))),
   c("Predicted", "Observed")))
mae(mat1[, 2], mat1[, 1])

# mat1[, 1] # observed values from column 1 of mat1
# mat1[, 2] # predicted values from column 2 of mat1


# using a data.frame of the numeric vectors pre1 and obs1
df1 <- data.frame(obs1, pre1)
mae(df1[, 2], df1[, 1])

# df1[, 1] # observed values from column 1 of df1
# df1[, 2] # predicted values from column 2 of df1


library("data.table")

# using a data.table of the numeric vectors pre1 and obs1
df2 <- data.table(obs1, pre1)
mae(df2[, 2, with = FALSE][[1]], df2[, 1, with = FALSE][[1]])

# df2[, 1, with = FALSE][[1]] # observed values from column 1 of df2
# df2[, 2, with = FALSE][[1]] # predicted values from column 2 of df2