Working with Outliers
An outlier is a data point that is significantly different from the remaining data. Statistical parameters such as the mean and variance are sensitive to outliers. Outliers may also affect the performance of some machine learning models, such as linear regression models. In these cases, we may want to remove or engineer the outliers in our variables.
How can we engineer outliers? One way to handle outliers is to perform variable discretization with any of the techniques we covered in Chapter 4, Performing Variable Discretization. With discretization, the outliers will fall in the lower or upper intervals and, therefore, will be treated as the remaining lower or higher values of the variable. An alternative way to handle outliers is to assume that the information is missing, treat the outliers together with the remaining missing data, and carry out any of the missing imputation techniques described in Chapter 1, Imputing Missing Data. We can also remove observations...