The following is a classwise evaluation index list for each target class, \(i\). Weighted averages of evaluation indices are subsequently computed wherein the weight of a target class is the proportion of the occurrences of the class in the actual population.

Evaluation Index

Type

Description

true_positive_<i> (**)

int

Number of samples determined as positive for each target class, \(i\), correctly (TP).

false_positive_<i> (**)

int

Number of samples determined as positive for each target class, \(i\), incorrectly (FP).

true_negative_<i> (**)

int

Number of samples determined as negative for each target class, \(i\), correctly (TN).

false_negative_<i> (**)

int

Number of samples determined as negative for each target class, \(i\), incorrectly (FN).

accuracy_<i>

float

Proportion of true results for each target class, \(i\), in the population as shown below:

\(\frac{\mbox{TP}_{i} + \mbox{TN}_{i}}{\mbox{TP}_{i} + \mbox{FP}_{i} + \mbox{TN}_{i} + \mbox{FN}_{i}}\)

classification_error_<i>

float

Proportion of false results for each target class, \(i\), in the population as shown below:

\(\frac{\mbox{FP}_{i} + \mbox{FN}_{i}}{\mbox{TP}_{i} + \mbox{FP}_{i} + \mbox{TN}_{i} + \mbox{FN}}_{i} = 1 - \mbox{accuracy}_{i}\)

precision_<i>

float

Proportion of the true_positive of each target class, \(i\), against all samples determined as positive as shown below:

\(\frac{\mbox{TP}_{i}}{\mbox{TP}_{i} + \mbox{FP}_{i}}\)

recall_<i>

float

Proportion of the true_positive of each target class, \(i\), against all the actual positive samples as shown below:

\(\frac{\mbox{TP}_{i}}{\mbox{TP}_{i} + \mbox{FN}_{i}}\)

specificity_<i>

float

Proportion of the true_negative of each target class, \(i\), against all the actual negative samples as shown below:

\(\frac{\mbox{TN}_{i}}{\mbox{TN}_{i} + \mbox{FP}_{i}}\)

false_positive_rate_<i>

float

Proportion of the false_positive of each target class, \(i\), against all the actual negative samples as shown below:

\(\frac{\mbox{FP}_{i}}{\mbox{TN}_{i} + \mbox{FP}_{i}} = 1 - \mbox{specificity}_{i}\)

false_negative_rate_<i>

float

Proportion of the false_negative of each target class, \(i\), against all the actual positive samples as shown below:

\(\frac{\mbox{FN}_{i}}{\mbox{TP}_{i} + \mbox{FN}_{i}} = 1 - \mbox{recall}_{i}\)

f_measure_<i>

float

Harmonic mean of precision and recall of each target class, \(i\), as shown below:

\(\frac{2 \times \mbox{precision}_{i} \times \mbox{recall}_{i}}{\mbox{precision}_{i} + \mbox{recall}_{i}}\)

cf_<i>_<j> (**)

int

Confusion matrix values that show the number of actual class, \(i\), values predicted as \(j\).
There are \(\mbox{num_target_classes}^{2}\) cf index values for every evaluation.
  • (**) Weighted average is not computed for this index.