Precision, Recall & F1-Score

#Day2ofML

Precision, recall are mostly used in information retrieval.

For binary classification

Precision = Of all the points, the model predicted to be +ve, what percentage of them are actually +ve

TP/TP+FP

Recall = Of all the points which actually belong to class +ve, how many our model detected to be +ve.

TP/TP+FN

• We always wants our precision and recall to be high
• Precision and recall lies between 0 and 1

If we want to combine both precision and recall and get it as one measure, we can get it as F1-score

F1-Score = harmonic mean of precision and recall

= [avg (inv-recall, inv-precision)]**-1

= 2*(precision*recall)/precision+recall