A New Approach to Outlier Detection

For many data mining applications, finding the rare instances or the outliers is more interesting than finding the common patterns. At present, many automated outlier detection methods are available, however, most of those are limited by assumptions of a distribution or require upper and lower predefined boundaries in which the data should exist. Whereas a distribution is often unknown, and enough information may not exist about a set of data to be able to determine reliable upper and lower boundaries. For these cases, a new dissimilarity function was defined, which can be viewed as fitness function of genetic algorithm, and a GA-based outlier detection method was formed in this paper. This method allows for detection of multiple outliers, not just one at a time. The illustrations show that the improved approach can automatically detect outliers, and performs better than GLOF approach.