Detecting outliers

An important aspect of creating training samples for supervised classification is to eliminate outliers in the sample data. An outlier is a data point that differs significantly from the remaining data. Outliers can arise due to factors such as human error, or from natural deviations in populations.

Detection and removal of outliers can eliminate their contaminating effect on the data set and purify the data for classification. Object Analyst provides three outlier-detection algorithms based on statistical analysis of the data:

Interquartile range (IQR)
Mean plus or minus three standard deviations
Mean plus or minus two standard deviations

In the analysis, the attributes you select are analyzed to identify potential outliers.