Abstract
KNN-Based outlier detection over IoT streaming data is a fundamental problem, which has many applications. However, due to its computational complexity, existing efforts cannot efficiently work in the IoT streaming data. In this paper, we propose a novel framework named GAAOD (Grid-based Approximate Average Outlier Detection) to support KNN-Based outlier detection over IoT streaming data. Firstly, GAAOD introduces a grid-based index to manage summary information of streaming data. It can self-adaptively adjust the resolution of cells, and achieve the goal of efficiently filtering objects that almost cannot become outliers. Secondly, GAAOD uses a min-heap-based algorithm to compute the distance upper-/lower-bound between objects and their k -th nearest neighbors respectively. Thirdly, GAAOD utilizes a k -skyband based algorithm to maintain outliers and candidate outliers. Theoretical analysis and experimental results verify the efficiency and accuracy of GAAOD.
Highlights
With the development of information science and technology [1]–[8], outlier detection [9] becomes more and more important
We study the problem of KNN-Based outlier detection over IoT streaming data
In this paper, we study the problem of outlier detection over IoT streaming data(short for streaming data)
Summary
With the development of information science and technology [1]–[8], outlier detection [9] becomes more and more important. Kontaki et al [22] propose a micro-clustering based algorithm named MCOD Their key idea is maintaining the last k arrived neighbours for each object, using maintenance results to evaluate which objects have chance to become outliers or cannot become outliers before they expire from the window. These algorithms only need to maintain the last k arrived neighbours for a small number of objects in the window These efforts only can support threshold-based outlier detection. Its computational complexity is high, and can not efficiently work under IoT streaming data environments To solve this problem, this paper studies the problem of approximate neighbour-based outlier detection over sliding window.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have