This basically means that it doesn't work at all. Accuracy of a test doesn't even mean anything in a vacuum, you have to know how common or rare the thing you're trying to detect is, so you can get actual false positive and false negative rates.
For example, let's say I have a facial recognition system that is 99% accurate. I'm looking for Osama Bin Laden. I deploy the facial recognition system on the new york subway and 100,000 people go through it. With a 1% error rate, 1,000 people are flagged as being Osama Bin Laden. The 99% accurate system in the lab is, at most, 0.1% accurate in the field.
>78%-85% Accurate
This basically means that it doesn't work at all. Accuracy of a test doesn't even mean anything in a vacuum, you have to know how common or rare the thing you're trying to detect is, so you can get actual false positive and false negative rates.
For example, let's say I have a facial recognition system that is 99% accurate. I'm looking for Osama Bin Laden. I deploy the facial recognition system on the new york subway and 100,000 people go through it. With a 1% error rate, 1,000 people are flagged as being Osama Bin Laden. The 99% accurate system in the lab is, at most, 0.1% accurate in the field.