The detection of outliers is a task in data mining and machine learning and it’s an important step in data preprocessing. In this paper, in order to detect proximity-based outliers, a non-parametric method is proposed called NPOD. The proposed method is a combination of More
The detection of outliers is a task in data mining and machine learning and it’s an important step in data preprocessing. In this paper, in order to detect proximity-based outliers, a non-parametric method is proposed called NPOD. The proposed method is a combination of distance-based and density-based methods and has the ability to detect outliers in both local and global scenarios. This method does not require to determine any parameters of neighborhood radius, the threshold of existing points in the neighborhood radius, and the nearest neighbor parameters. In order to detect outliers, a new method of scoring is presented. Experimental results on the UCI datasets show that this algorithm, in spite of being non-parametric, has comparable results with previous methods. Also in some cases, it has the best performance.
Manuscript profile
