ml_test {mltest}

multi-class classifier evaluation metrics based on a confusion matrix (contingency table)

Description

Calculates multi-class classification evaluation metrics: balanced.accuracy, balanced accuracy (balanced.accuracy), diagnostic odds ratio (DOR), error rate (error.rate), F.beta (F0.5, F1 (F-measure, F-score), F2 with where beta is 0.5, 1 and 2 respectively), false positive rate (FPR), false negative rate (FNR), false omission rate ((FOR)), false discovery rate (FDR), geometric mean (geometric.mean), Jaccard, positive likelihood ratio (p+, LR(+) or simply L), negative likelihood ratio (p-, LR(-) or simply lambda), Matthews corellation coefficient (MCC), markedness (MK), negative predictive value (NPV), optimization precision OP, precision, recall (sensitivity), specificity and finally Youden's index. The function calculates the aforementioned metrics from a confusion matrix (contingency matrix) where TP, TN, FP FN are abbreviations for true positives, true negatives, false positives and false negatives respectively.

Usage

ml_test(predicted, true, output.as.table = FALSE)

Arguments

predicted

class labels predicted by the classifier model (a set of classes convertible into type factor with levels representing labels)

true

true class labels (a set of classes convertible into type factor of the same length and with the same levels as predicted)

output.as.table

the function returns all metrics except for accuracy and error.rate in a tabular format if this argument is set to TRUE

Value

the function returns a list of following metrics:

\strong{accuracy}

calculated as: (TP+TN) / (TP+FP+TN+FN) (doesn't show up when output.as.table = TRUE)

\strong{balanced.accuracy}

calculated as: (TP / (TP+FN)+TN / (TN+FP)) / 2 = (recall+specificity) / 2

\strong{DOR}

calculated as: TP*TN / (FP*FN) = L / lambda

\strong{error.rate}

calculated as: (FP+FN) / (TP+TN+FP+FN) = 1-accuracy (doesn't show up when output.as.table = TRUE)

\strong{F0.5}

calculated as: 1.25*(recall*precision/(0.25*precision+recall))

\strong{F1}

calculated as: 2*(precision*recall / (precision+recall))

\strong{F2}

calculated as: 5*(precision*recall / (4*precision+recall))

\strong{FDR}

calculated as: 1-precision

\strong{FNR}

calculated as: 1-recall

\strong{FOR}

calculated as: 1-NPV

\strong{FPR}

calculated as: 1-specificity

\strong{geometric.mean}

calculated as: (recall*specificity)^0.5

\strong{Jaccard}

calculated as: TP / (TP+FP+FN)

\strong{L}

calculated as: recall / (1-specificity)

\strong{lambda}

calculated as: (1-recall) / (specificity)

\strong{MCC}

calculated as: (TP*TN-FP*FN) / (((TP+FP)*(TP+FN)*(TN+FP)*(TN+FN))^0.5)

\strong{MK}

calculated as: precision + NPV - 1

\strong{NPV}

calculated as: TN / (TN+FN)

\strong{OP}

calculated as: accuracy - |recall-specificity| / (recall+specificity)

\strong{precision}

calculated as: TP / (TP+FP)

\strong{recall}

calculated as: TP / (TP+FN)

\strong{specificity}

calculated as: TN / (TN+FP)

\strong{Youden}

calculated as: recall+specificity-1

Author(s)

G. Dudnik

References

1. Sasaki Y. (2007). The truth of the F-measure.:1–5. https://www.researchgate.net/publication/268185911_The_truth_of_the_F-measure.

2. Powers DMW. (2011). Evaluation: from Precision, Recall and F-measure to ROC, Informedness, Markedness & Correlation. Arch Geschwulstforsch. 2(1):37–63. https://www.researchgate.net/publication/313610493_Evaluation_From_precision_recall_and_fmeasure_to_roc_informedness_markedness_and_correlation.

3. Bekkar M, Djemaa HK, Alitouche TA. (2013). Evaluation Measures for Models Assessment over Imbalanced Data Sets. J Inf Eng Appl. 3(10):27–38. https://www.iiste.org/Journals/index.php/JIEA/article/view/7633.

4. Jeni LA, Cohn JF, De La Torre F. (2013). Facing Imbalanced Data Recommendations for the Use of Performance Metrics. Conference on Affective Computing and Intelligent Interaction. IEEE. p. 245–51. http://ieeexplore.ieee.org/document/6681438/.

5. López V, Fernández A, García S, Palade V, Herrera F. (2013). An insight into classification with imbalanced data: Empirical results and current trends on using data intrinsic characteristics. Inf Sci. 250:113–41. http://dx.doi.org/10.1016/j.ins.2013.07.007.

6. Tharwat A. (2018). Classification assessment methods. Appl Comput Informatics . https://linkinghub.elsevier.com/retrieve/pii/S2210832718301546.

Examples

library(mltest)

# class labels ("cat, "dog" and "rat") predicted by the classifier model
predicted_labels <- as.factor(c("dog", "cat", "dog", "rat", "rat"))

# true labels (test set)
true_labels <- as.factor(c("dog", "cat", "dog", "rat", "dog"))

classifier_metrics <- ml_test(predicted_labels, true_labels, output.as.table = FALSE)

# overall classification accuracy
accuracy <- classifier_metrics$accuracy

# F1-measures for classes "cat", "dog" and "rat"
F1 <- classifier_metrics$F1

# tabular view of the metrics (except for 'accuracy' and 'error.rate')
classifier_metrics <- ml_test(predicted_labels, true_labels, output.as.table = TRUE)