A new outlier detection method based on OPTICS

OPTICS is a density-based clustering method that can address point sets with different densities; however, the outlier detection capability of OPTICS is limited by several factors, such as different parameter and different point set shapes. Therefore, an outlier detection method based on OPTICS was proposed, known as OD-OPTICS, which adds a pre-processing process and modifies the CD computing method. First, the Radius Filtration Strategy, which provides the key radii, is performed; and the crucial distances of the point set are reflected. Then, for the purpose of filtrating invalid radii and selecting the most appropriate one, the Covering Space Model was proposed. With the influence of the three rules that we established, the basic distance between two neighbouring points can be calculated by the covering space. Moreover, the CD calculation was optimized such that it could magnify the difference value between normal points and outliers. In the experiment part, the preprocessing details were demonstrated and the validity of OD-OPTICS on public point sets were tested; to verify the optimization and detectability of OD-OPTICS, the proposed method is compared to OPTICS and four other typical methods. The results demonstrated that the detection performance of OD-OPTICS is superior to that of OPTICS.