Premium
An Anomaly Detection Method for Individual Services on a Web‐Based System by Selection of Dummy Variables in Multiple Regression
Author(s) -
Tsuda Yuki,
Samejima Masaki,
Akiyoshi Masanori,
Komoda Norihisa,
Yoshino Matsuki
Publication year - 2014
Publication title -
electronics and communications in japan
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.131
H-Index - 13
eISSN - 1942-9541
pISSN - 1942-9533
DOI - 10.1002/ecj.11660
Subject(s) - anomaly detection , anomaly (physics) , computer science , linear regression , regression analysis , service (business) , data mining , precision and recall , web service , regression , resource (disambiguation) , selection (genetic algorithm) , statistics , mathematics , machine learning , computer network , world wide web , physics , economy , economics , condensed matter physics
SUMMARY This paper addresses the detection of anomalies of individual service from their total resource usage on a web‐based system. Because the total resource usage is a linear combination of the number of accesses to each service, multiple regression analysis can be used to estimate the resource usage per access to each service in the form of regression coefficients. However, the regression coefficients differ from the resource usage per access of the services, due to unstable resource usage per access. We propose a method based on the multiple correlation coefficient R to identify anomaly times and anomaly services. The proposed method identifies an anomaly time when the R value is decreased; it identifies an anomaly service by judging whether R value has increased after the selection of the dummy variable. Experimental results show that the proposed method can identify all anomaly times, and that it improves the precision rate and recall rate of anomaly service detection by at least 20%.