When you discover that there are outliers, the obvious follow-up question is how to handle them. I'm a fan of simple and efficient solutions, so my favorite way is to simply exclude the corresponding records. The problem with exclusion is that there may actually be some degree of truth to these values. Of course, a person will not be 9,999.9 m large, but when you look for example at revenue of individual customers, there may actually be a single person who spent 200,000.00 EUR with your company in the last year, even though the average customer only spends 200.00 EUR with you. But even if the 200,000.00 EUR are accurate, including it in training data for a model may hurt you, depending on whether your model is sensitive to outliers and what metric you use for assessing model quality. Unfortunately, there are also situations where discarding a record is not an option. If you are using a model to predict something about this particular customer, you need to be able to handle the outlier in some way.