Evaluation Basics: True Positive, False Positive, True Negative, False Negative

Hello students, welcome to Kwickprep. An AI test says a person does not have a disease, but they actually do. Is that a small mistake or a dangerous one? Today we learn to check an AI model's answers using true positive, false positive, true negative and false negative.

First, a new term. Evaluation means checking how good a model is. It is the fifth stage of the AI project cycle, after modelling. We test the model on new data it has never seen before. Then we compare what the model predicted with what was really true.

We need two words for every case. The prediction is what the model says. The reality is what is actually true, which we find out later, for example from a lab report. Also, in AI, positive simply means yes and negative means no, not good or bad.

Prediction can be yes or no, and reality can be yes or no, so there are four boxes in a two by two grid. In the top row, the model predicted yes, so the reality decides true positive or false positive. In the bottom row, the model predicted no, which gives false negative or true negative.

Here is an easy trick to read these names. The second word, positive or negative, tells what the model predicted. The first word, true or false, tells whether the model was right. True means right, and false means wrong.

Let us use an AI model that tests for dengue. A true positive means the model says sick, and the person really is sick. A true negative means the model says healthy, and the person really is healthy. Both of these are correct predictions, so the model did well.

Now the two mistakes. A false positive means the model says sick, but the person is actually healthy, a false alarm. A false negative means the model says healthy, but the person is actually sick, a missed case. Both of these are wrong predictions.

Pause and predict. The model says the patient has no dengue. But the lab report says the patient has dengue. Which box is it? The model predicted negative and it was wrong, so it is a false negative.

Both mistakes are wrong, but are they equally bad? A false positive means a healthy person gets an extra test, which costs some worry and money, but the second test clears them. A false negative means a sick person goes home without treatment, so the illness can get worse and even spread to others.

So which mistake is worse depends on the problem. For a disease test, a false negative is worse, because a sick person is missed. For an email spam filter, a false positive is worse, because an important email, like an exam admit card, lands in spam. So always ask, which mistake costs more in this problem?

Now let us count outcomes from a small table of six patients. Asha is predicted sick and is sick. Babu is predicted sick but is healthy. Chirag is predicted healthy and is healthy. Divya is predicted healthy but is sick. Esha is predicted sick and is sick. Farhan is predicted healthy and is healthy.

Let us sort each patient into a box. Asha and Esha were predicted sick and are sick, so there are two true positives. Babu was predicted sick but is healthy, so one false positive. Chirag and Farhan were predicted healthy and are healthy, so two true negatives. Divya was predicted healthy but is sick, so one false negative.

Always check your counting in exams. Add all four boxes, two plus one plus two plus one, and you get six, the number of patients. The correct predictions are true positives plus true negatives, which is four. The wrong predictions are false positives plus false negatives, which is two.

Let us revise what we learned today. Evaluation compares the model's prediction with reality. There are four boxes in a two by two grid: true positive, false positive, true negative and false negative. The second word tells the prediction, and the first word tells if it was right. In a disease test, a false negative is often worse than a false positive. And the four counts always add up to the total number of cases.