A New Outlier Detection Method Based On PageRank Algorithm And Its Application

In order to find out the useful information, we usually need to detect outliers in given images and databases. In these cases, outlier detection is essential. Besides these, outlier detection can be also used in machine vision, pattern recognition, curve-fitting and so on. Obviously, outlier detection is a classical issue with comprehensive applications. So far in the field of outlier detection, many scholars have advanced a number of effective detecting algorithms. But all of these algorithms have kinds of limitations, such as a detecting radius must be given. These limitations tend to make the detecting results with a large amount of human subjectivity.In order to overcome those shortcomings mentioned above, in this paper, we propose a new outlier detection method based on PageRank algorithm in which outliers are divided into three categories and we use different outlier detecting strategies for different categories. The first kind of outliers is detected by a given domain radius which is given by the algorithm automatically. The second kind of outliers is detected by introduction of the isolated degree. Finally, the third kind of outliers is detected on the ground of the PageRank algorithm. The whole process is called triple sifting or triple detecting.Simulation experiments of the algorithm are carried out with the Matlab. The comparison results between the algorithm based on PageRank and the algorithm based on density suggest that the outlier detection algorithm based on PageRank is an effective algorithm.