Accuracy indices

{lvmisc} contains a group of useful functions to compute basic indices of accuracy. These functions can be divided in those which compute element-wise values and those which compute average values:

  • Element-wise:
    • error()
    • error_abs()
    • error_pct()
    • error_abs_pct()
    • error_sqr()
  • Average:
    • mean_error()
    • mean_error_abs()
    • mean_error_pct()
    • mean_error_abs_pct()
    • mean_error_sqr()
    • mean_error_sqr_root()
    • bias()
    • loa()

You may notice that the majority of these functions have common prefixes (error_ and mean_error_), intended to facilitate the use, as most text editors have an auto-complete feature. Also all of the accuracy indices functions take actual and predicted as arguments, and the functions that return average values have na.rm = TRUE in addition.

Let’s now see how each function computes its results

Element-wise

Error: error()

It simply subtracts the predicted from the actual values.

Formula: ai − pi

Absolute error: error_abs()

It returns the absolute values of the error() function.

Formula: |ai − pi|

Percent error: error_pct()

Divides the error by the actual values.

Formula: $$\frac{a_i - p_i}{a_i}\cdot100$$

Absolute percent error: error_abs_pct()

Returns the absolute values of the error_pct() function.

Formula: $$\frac{|a_i - p_i|}{|a_i|}\cdot100$$

Squared error: error_sqr()

It squares the values of the error() function.

Formula: (ai − pi)2

Average

Mean error: mean_error()

It is the average of the error.

Formula: $$\frac{1}{N}\sum_{i = 1}^{N}(a_i - p_i)$$

Mean absolute error: mean_error_abs()

Computes the average of the absolute error.

Formula: $$\frac{1}{N}\sum_{i = 1}^{N}|a_i - p_i|$$

Mean percent error: mean_error_pct()

The average of the percent error.

Formula: $$\frac{1}{N}\sum_{i = 1}^{N}\frac{a_i - p_i}{a_i}\cdot100$$

Mean absolute percent error: mean_error_abs_pct()

It is the average of the absolute percent error.

Formula: $$\frac{1}{N}\sum_{i = 1}^{N}\frac{|a_i - p_i|}{|a_i|}\cdot100$$

Mean squared error: mean_error_sqr()

Averages the mean squared error.

Formula: $$\frac{1}{N}\sum_{i = 1}^{N}(a_i - p_i)^2$$

Root mean squared error: mean_error_sqr_root()

It takes the square root of the mean squared error.

Formula: $$\sqrt{\frac{1}{N}\sum_{i = 1}^{N}(a_i - p_i)^2}$$

Bias: bias()

Alias to mean_error().

Limits of agreement: loa()

Formula: bias ± 1.96σ

Where σ is the standard deviation.