Metric
Formula
Interpretation
Pros
Cons
Accuracy
Total Number of PredictionsNumber
Proportion of correctly
Easy
of Correct
classified
to understand.
Predictions
instances.
Can be misleading with imbalanced datasets.
Precision
True Positives (TP)
Proportion
+ False Positives
of predicted
Useful
(FP)True
positives
whenPositives
thethat
cost
Doesn't
are
of
(TP)
actually
false
consider
positives
positive.
falseisnegatives.
high.
Recall (Sensitivity)True Positives (TP)
Proportion
+ False Negatives
of actual
Useful
positives
(FN)True
whenthat
the
Positives
are
cost
Doesn't
correctly
of(TP)
false
consider
negatives
identified.
false ispositives.
high.
F1-Score
2×Precision+RecallPrecision×Recall
Harmonic mean ofBalances
precision1and
precision
recall.
Can and
be harder
recall, to
useful
interpret
for imbalanced
than accuracy.
datasets.
Area Under the ROC
AreaCurve
under(AUC-ROC)
the Receiver
Measures
Operating
the ability
Provides
Characteristic
of the a
classifier
single
(ROC)
Doesn't
measure
to distinguish
curve.
give
of overall
insight
between
performance
into classes.
the type across
of errors
different
being made.
thresholds, robust to class imbalan
Sources
https://github.com/karthigamuthuraj/MachineLearning