题目: Isolation Forest for Anomaly Detection 报告人：Dr. Kai Ming Ting Monash University, Australia
时间：14:30-15:30 Dec.4 地点：蒙民伟楼404 摘要:
Most existing approaches to anomaly detection construct a profile of normal in
stances, then identify instances that do not conform to the normal profile as
anomalies. This paper proposes a fundamentally different method that explicitl
y isolates anomalies instead of profiles normal points. To our best knowledge,
the concept of isolation has not been explored in the current literature. The
use of isolation enables the proposed method, Isolation Forest (iForest), to
exploit sub-sampling to an extent that is not feasible in existing methods, cr
eating an algorithm which has a linear time complexity with a low constant and
a low memory requirement. This has no parallel in anomaly detection methods t
hat we have surveyed.
Our empirical evaluation shows that iForest performs favourably to ORCA (a nea
r-linear time complexity distance-based method), LOF (a density-based method)
and Random Forests in terms of AUC and processing time, and especially in larg
e data sets. iForest also works well in high dimensional problems which have a
large number of irrelevant attributes, and in situations where training set d
oes not contain any anomalies.
This is a joint work with Tony Liu and Zhi-Hua Zhou.
简历:
After receiving his PhD from the University of Sydney, Australia, Dr. Ting had
worked at the University of Waikato, New Zealand and Deakin University, Austr
alia. He joins Monash University since 2001. He is the one of the founding mem
bers of Centre for Research in Intelligent Systems at Monash University. Dr Ti
ng was one of the three program co-chairs for The Twelfth Pacific-Asia Confere
nce on Knowledge Discovery and Data Mining recently held in Osaka, Japan in Ma
y 2008, and has been a program committee member for 19 international conferenc
es in the field since 2003, including IEEE International Conference on Data Mi
ning, International Conference on Machine Learning, and ACM SIGKDD Internation
al Conference on Knowledge Discovery and Data Mining. His current research int
erests are in the areas of ensemble methods, anomaly detection, cost-sensitive
learning, model evaluation methods, swarm-based clustering, data mining and m
achine learning in general. His recent papers can be found at http://www.gscit
.monash.edu.au/~kmting/publications.html |
