Identification and Processing of Outliers

Abstract

When analyzing real data sets, observations different from the majority of the data are sometimes found. These observations are usually called outliers and can be defined as individual data values that are numerically distant from the rest of the sample, thus masking its probability distribution. Outliers require special attention because they can have a significant impact in the concrete strength estimation process and because they may signal the presence of a different concrete population that deserves a separate assessment. The two-step process involved in an outlier analysis (outlier identification and outlier handling) is presented, discussing several statistical methodologies that are available for its implementation. To illustrate the application of an outlier analysis, examples involving univariate and multivariate datasets are presented. Several statistical methodologies are implemented for outlier identification, while outlier handling is illustrated by using robust statistics, i.e. outlier accommodation approaches that reduce the effect of existing outliers on the outcomes of statistical analyses of the data.