When you buy through affiliate links in our content, we may earn a commission at no extra cost to you. Learn What Does Novelty Detection Mean?
Novelty detection is a statistical method used to determine new or unknown data and determining if these new data are within the norm (inlier) or outside of it (outlier).
A novel in this case means unusual, data that are new and doe not occur regularly or are simply different from the others. Novelty is applied in different fields which need to detect anomalies in their regular operations such as detecting network intrusions, hacking, jet engine failure, machine learning, and many more.
In fraud detection, for example, credit card companies monitor the spending habits of a user and when there is a deviation from those habits, they immediately call the user to ask whether the purchase was made legitimately or if the card was lost or stolen.
Techopedia Explains Novelty Detection
Novelty detection is one of the fundamental requirements for a proper classification system and in machine learning. In machine learning systems, not all possibilities can be inputted during training, so there will always be new kinds of data and possibilities that will arise in the future, basically inputs that differ from those that are regularly received or seen.
In fault and fraud detection for example, the system is trained to detect data that have been underrepresented or have not been seen at all since these are potential faults, and in medical data systems, this could represent disease.
For pure novelty detection systems, the network is trained on the negative examples, then only detects inputs that do not fit into this model as novel class.
Recognizing that an input differs from previous inputs is a very important and useful ability for learning systems. That would mean that the system is able to really learn and not just react to previous inputs and programming.
In the case of animals and humans, we exercise novelty detection all the time; that is the ability to distinguish objects from other objects. For example, we see a plain white wall and then see a speck moving on its surface, we immediately de-associate it from the wall remarking that it is a different object, probably an insect.
