An efficient outlier detection method based on distance ratio of k-nearest neighbors
Reza Heydari Gharaei, Rasoul Sharify, Hossein Nezamabadi–pour · 2022
Outlier detection is an important topic, which has been investigated in different theoretical and applied sciences. So far, numerous methods and algorithms have been proposed for outlier detection; each of these methods has found a use in some fields and has advantages and disadvantages. The present study presents an efficient method for detecting outliers based on the distance ratio of the k-nearest neighbors. In this method, both criteria of density and distance between objects have been included. Contrary to a large number of the previous methods, the density of each object is measured in relation to each neighbor independently, and after considering the distance criterion, the outlierness score of each object is measured collectively. Another feature of this method is its resistance to the k parameter. The proposed method has been evaluated in two-dimensional synthetic and multi-dimensional real datasets and compared with other significant algorithms in this field. The results of the experiments proved the efficiency of the proposed method.