CLASSIFICATION REPORT

You might have already come across some classification report like this

Let's discuss this report in detail

Classes

0, 1, 2 are the three different classes

Support

Support is nothing but the number of test samples available for testing

In this example
Total test samples available = 100
Test samples that belong to class 0 = 39
Test samples that belong to class 1 = 27
Test samples that belong to class 2 = 34

Precision

Suppose with the help of the learning algorithm, you have predicted that 31 out of those 100 test samples belong to class 0

Only 11 of those 31 samples you predicted that belong to class 0 are actually belongs to class 0

So those 11 are considered as true positives because they are correctly classified that they belong to class 0 and the rest are considered as false positives because we predicted them that they belong to class 0 but they don’t belong to class 0

True Positives = TP = 11
False Positives = FP = 20
Precision = (TP)/(Total samples that were predicted as positives)
Precision = (TP)/(TP + FP)
Precision = 11/31 = 0.35

Recall

There are 39 samples that belong to class 0

Suppose with the help of the learning algorithm, we have predicted that 11 out of those 39 samples belong to class 0

11 → True Positives

28 of them are predicted that they don’t belong to class 0 but they actually belong to class 0

28 → False Negatives

True Positives = 11
False Negatives = 28
Recall = TP/Total samples that were actually positive
Recall = TP/(TP + FN)
Recall = 11/39 = 0.28

F1-Score

F1-score provides a relationship between precision and recall. So this helps us in quantifying our learning algorithm better and brings balance between precision and recall

f1-score = (2 * precision * recall) / (precision + recall)

If either precision or recall is pretty small, f1-score is going to be pretty small

precision of class 0 = 0.35
recall of class 0 = 0.28
f1-score = (2 * 0.35 * 0.28) / (0.35 + 0.28) = 0.31

Accuracy

Total number of samples that are classified into the right correct classes with the help of the learning algorithm

Total test samples → 100

Samples that are classified into the correct class with the help of our learning algorithm → 30

accuracy = 30/100 = 0.3

Macro Average

Compute the average without considering the proportion

precision_class_0 = 0.35
precision_class_1 = 0.24
precision_class_2 = 0.33
macro-avg precision = (0.35 + 0.24 + 0.33)/3

Weighted Average

Compute the average considering the proportion

precision_class_0 = 0.35
precision_class_1 = 0.24
precision_class_2 = 0.33
samples_class_0 = 39
samples_class_1 = 27
samples_class_2 = 34
total_samples = 100
proportion_class_0 = 39/100 = 0.39
proportion_class_1 = 27/100 = 0.27
proportion_class_2 = 34/100 = 0.34
w-avg precision = ((0.35 * 0.39) + (0.24 * 0.27) + (0.33 * 0.34))

Similarly the calculations of macro-avg recall, weighted-avg recall, etc. can be done