An Approach for Discovering Rare High-Density Outlier Cluster

Outlier detection is one of the common tasks in engineering applications. Its aim is to identify uncommon records from large amounts of data. Many efforts have been done to reach this goal. But different approach should be introduced according to different application area. In some engineering cases, much frequency uncommon records may appear in a short time. In this paper, a new method is put forward to identify these uncommon records. This method is constructed based on density-based outlier discovering method. In this method, a ratio factor is used to illustrate the high-density percentage of all records. If the ratio is under a threshold the high-density records would be considered as outlier, otherwise these records would be considered as normal. Experiments show this method could effectively discovery these rare high-density data.