Linear Models for Outlier Detection

Abstract

The different dimensions in real data sets are highly correlated with one another. This is because the different attributes are usually generated by the same underlying process in closely related ways. In the classical statistics literature, this is referred to as regression modeling, a parametric form of correlation analysis. Some forms of correlation analysis attempt to predict individual attribute values from others, whereas other forms summarize the entire data in the form of latent variables. An example of the latter is the method of principal component analysis. Both forms of modeling can be very useful in different scenarios of outlier analysis. This chapter will discuss the different methods for using linear correlation analysis for outlier detection.