Why is everyone afraid of false negatives?

Any machine learning classification model is evaluated on error parameters to understand the models working in real-time. The main aim of developing a model is to ensure reliability and to obtain the right predictions from the model. So from any developed model, false negatives are a very crucial error parameter yielded. Higher false negatives from the model result in a dangerous or costly error termed a Type-2 error. Through this article let us try to understand why everyone is afraid of false negatives.

Table of Contents

1. What is a false negative?
2. Risk of false negatives - some use cases
3. Summary

What is a false negative?

The above image is the pictorial representation of the various parameters that can be obtained and evaluated from the confusion matrix. So now let us try to understand false negatives in detail.

False-negative is the result produced by the model indicating the absence of a particular condition or result when it is actually present. The magnitude of false-negative varies according to the model's capability to correctly classify the instance in real-time. In statistical terms, the false negatives are referred to as Type-2 errors and these types of errors are known to yield faulty predictions from the models developed.

Are you looking for a complete repository of Python libraries used in data science, check out here.

So False negatives will be responsible for accounting for very serious consequences from the models developed if it's not addressed or if the magnitude of False negatives is not in the considerable range. So in the next section of the article let us try to understand why everyone is afraid of false negatives through some examples.

Risk of false negatives - some use cases

Let us try to understand how false negatives will be responsible for faulty predictions and why everyone is afraid of false negatives. We will try to understand this risk with some use cases as given below.

Disease prediction

Say suppose we have developed a model to predict whether the patient is having cancer or not. So once the model is developed using the data available the model is evaluated for various terms. The magnitude of False negatives can be easily evaluated through the confusion matrix. But if we evaluate the significance of False negatives statistically the null hypothesis states would state that the patient is not having cancer. Keeping this in mind if the model developed predicts that the patient is not having cancer and in actuality, if the patient has cancer, it would leave the patient undiagnosed for the disease.

Survival prediction

For example, consider we are formulating a model for survival prediction which is a very sensitive model and the model has to come up with predictions without any room for errors. The data and the parameters for this type of modelling are very crucial. Say suppose subjectively the survival of a patient is not possible and the model is saying that the patient would survive, the model is responsible for yielding a wrong prediction. This wrong prediction would be responsible for not diagnosing the patient and helping them survive.

Covid Test Result

Let us consider that a patient goes for a covid test and he obtains the report. The model that will be responsible for generating the report has to be very accurate. Let us consider that in actuality if the individual is tested positive for covid and if the report generated is negative this is a very serious consequence as the individual may spread the infection to a large set of population or to the people in his contact. So this type of error will be classified as a false negative or Type-2 error.

Fake News Prediction

Let us consider that we are building a news classification model where the news has to be classified correctly into genuine and fake news. Say suppose if, in reality, the news is genuine and the model developed has classified the news as fake the news would be unread by the larger population in the society. This may lead to the reach of a very genuine or real news update to the people. So this scenario can be termed a false-negative and this, in turn, results in the Type-2 error.

Summary

False negatives are one of the very important and crucial evaluation metrics of various models being developed. The model developed for crucial predictions like some of them explained in this article has to be very robust for false negatives as it would lead to serious consequences or faulty predictions. So this is why everyone is afraid of false negatives and they try to address and evaluate the false negatives on priority before obtaining required predictions from the model.