On New Approach in Mining Outlier

Abstract

In this paper, a new outliers detection technique called RSetOF is proposed. The proposed technique is based on rough set theory that has significant contribution in classification rules mining. Several concepts on theoretical aspect of rough set such as Indiscernibility relation and reduct computation are employed in this study. RSetOF is a new measure for the outliers factors that based on rough set theory. By employing this factor, a new formulation of detecting outlier is established. A comparative study is carried out to examine the performance of the RSetOF. The RSetOF is compared with other three recent outliers detection techniques; RNN, CBLOF and FPOF in terms of the measure used and the ability to identify interesting outliers objects in the dataset. These techniques use the concept of outliers factor as in the RSetOF with different methodology. The measurement is presented in percentage on top-n ranking of detected outliers in anticipation of all assumed outliers in the dataset are found. A benchmark dataset for assessing the outlier detection techniques is used for this purpose. The experimental result shows that the proposed technique is competitive with other techniques and proven to be better in speed of detection. It brings potential as a new rough set based outliers detection technique