A novel approach to identify anomalies using rough sets

Most datasets contain objects whose attributes differ significantly as compared to those of other objects in the same dataset. Although initially disregarded as noise such objects are now defined as outliers and detecting them can be beneficial for applications such as detecting fraudulent financial transactions, or intruders. A major challenge to detect these outliers is the dimensionality and vastness of data. Rough sets can be used to clearly define the objects that need not be considered as outliers. As a result, all objects do not need to be processed while applying the outlier detection algorithm. This paper exploits a new methodology for detecting outliers using rough sets. This methodology has high potential since outliers have a low probability of being in the boundary region defined by the intersection of the lower and upper approximation, as compared to the lower approximation. This use of rough sets can be used to significantly reduce the computation time of existing algorithms.