What is Imputation?
In essence, imputation is simply replacing missing data with substituted values. Often, these values are simply taken from a random distribution to avoid bias.
Imputation is a fairly new field and because of this, many researchers are testing the methods to make imputation the most useful. Currently, the methods we have to implement imputation aren’t as effective as researchers believe they could be, and involve introducing bias or by decreasing the representative power of the results.
Applications in Artificial Intelligence
Sometimes (and unfortunately) the data we have to feed to computers is incomplete. Because this is unavoidable, computers do need to know how to handle incomplete data sets. If you teach a computer to simply ignore incomplete data, it will have less data to work with, making it less likely that it will come up with a useful answer. In some cases, AI can use former knowledge to complete the data, but currently this is not possible in all situations. So what’s the solution? Imputation. What the AI doesn’t know it fills in, often randomly, to be able to produce an answer.