Fig 1:  Look Alike World With No Outiler

Fig 2 : Look Alike World: but with An Outlier

Imagine a look alike world like the picture, everybody looks identical, no variance in appearance. If that was the case, the notion of outliers would not have been relevant.

However, even in such a look alike world, you may get an outlier, which exhibits 'off the average' behavior. Look at the bottom right picture in Fig 2, you'll see a face of a cat, certainly making itself an outlier.

So, intuitively, Outliers are a set of data points that are considerably dissimilar or inconsistent with the remainder of the data.

Are outliers always a nuisance ?

No. Although in some case, they may impede the inference process, in other cases, they may actually carry valuable information.

Outliers have interesting and useful application in Telecom, Credit Card Fraud Detection, Loan Approval, Pharmaceutical Research, Weather Prediction, Financial Applications and Customer Segmentation.