# Is it possible to estimate accuracy, precision and recall with the given data?

Background: I talked to my friend today and according to herm(him/her) I can calculate precision, recall and accuracy with the current information.

Total instances T: 19,532.

Instances belonging to class F: 8829.

What my classifier totally found, lets call it TS: 11,120.

The number of instances of TS belonging to F, lets call it FTS: 6622.

What is TP, TN, FP and FN here? From my understanding I get these, not sure if it is correct though.

1. (TP: 6622)
2. (TN: 11120-6622=4498)
3. (FP:11120-8829=2291)
• What does "What my classifier totally found" mean? Is it number of positives, i.e. class F, your classifier found out of 19532? – gunes May 7 '19 at 20:15
• @gunes Yes, out of 19,532 instances, my classifier classified 11,120 to belong to class TS. Out of 11,120 of those instances, 6622 belonged to a the desired class F. – sockevalley May 7 '19 at 20:18
• There should be two classes: F = positives, and F' = negatives. No class TS as I understand. So, 11,120 is classified as class F by your classifier. But, 6622 were really F. Isn't it? – gunes May 7 '19 at 20:35
• @gunes Yes. 6622 is really F class, sorry for the confusion. – sockevalley May 7 '19 at 20:40

I'll slightly round up the numbers for ease of notation.

1. Your TP, i.e. True Positives is correct.
2. TN (True Negatives) + FP (False Positive) = Total Negatives = $$19.5K-8.8K=10.7K$$. Plug-in the FP in (3) and get TN.
3. FP (False Positives) = We called Positive - True Positive = $$11.1K-6.6K\approx 4.5K$$ (i.e. your answer for TN is actually FP).

Precision, recall and accuracy can be calculated easily from these three.

• I became very confused by this answer. Did I get this correct? FP = 4.5K TP = 6.6K TN = 10.7k FN = incorrectly rejected..? – sockevalley May 7 '19 at 21:06
• Wasn't 4.5k FP and not 11.7K -4.5K? – sockevalley May 7 '19 at 21:31
• Cheers! How do you calculate the FN? – sockevalley May 8 '19 at 6:07
• FN is False Negatives = (We called Negative) - (True Negatives). TN is found in (2). We called 19.5K-11.1K negatives in total. – gunes May 8 '19 at 6:11
• +1. The Wikipedia page on sensitivity and specificity is long, but has a lot of very good information and examples. – Stephan Kolassa May 8 '19 at 6:37