ARTIFICIAL IMMUNITY-BASED CORRELATION SYSTEM
Guillermo Suarez-Tangil, Esther Palomar, Sergio Pastrana and Arturo Ribagorda
Department of Computer Science, University Carlos III of Madrid, Avda. Universidad 30, 28911 Madrid, Spain
Keywords:
Artificial immune system, Event correlation, Security event information management system, Intelligent rule
generation, Adaptive system.
Abstract:
Security information event management (SIEM) technologies focus on developing effective methods and tools
to assist network administrators during the whole network security management. Though there is a vast
number of novel initiatives and contributions in providing adaptiveness and intelligence in this research field,
there are still many problems that need be solved. In particular, event correlation are currently emerging as
an essential field to be optimized specially due to the widespread adoption of botnets to launch attacks. This
position paper explores the biological immune system’s characteristics of learning and memory to solve the
semi-automatic generation of event correlation rules by applying Artificial Immune Systems (AISs).
1 PROBLEM STATEMENT
Nowadays, network security management has to deal
with two main critical tasks: On the one hand, dif-
ferent security data sources (known as sensors), e.g.
intrusion detection systems (IDSs), firewalls, server
logs, to name a few produce high amounts of het-
erogeneous information, generally difficult to under-
stand. Moreover, attackers may use IDS-stimulators
to disguise their intrusions by hiding attacks into an
alert storm (Mutz et al., 2003). On the other hand,
the continuous evolution of attacks, specially recent
distributed, multi–step attacks, poses a challenge to
the network infrastructure as these attacks may be not
noticed when they are inspected separately.
To encounter these challenges, cooperation among
these sensors becomes essential for the former, whilst
event correlation appears as the best palliative for
the latter. Security Information and Event Manage-
ment (SIEM) systems, as a holistic solution, help to
gather, organize and correlate security network infor-
mation as well as reducing the amount of time spent
by security administrators. A typical SIEM architec-
ture facilitates experts to supervise the security status
of an organization. However, current SIEM systems
lack of an efficient mechanism to generate correlation
rules and cannot adaptively predict novel attacks ei-
ther (Anuar et al., 2010).
By providing an intelligent adaptability to the cor-
relation engine, we envision that time spent on de-
tecting zero-day attacks can be significantly reduced.
For instance, various works focused on the applica-
tion of Artificial Intelligence (AI) techniques to par-
tially optimize IDSs. For example, neural networks
(NN), widely used for optimizing classification prob-
lems (Ripley, 1994) have been also applied to im-
prove misuse filtering and malicious pattern recogni-
tion (Lippmann and Cunningham, 2000). Moreover,
Evolutionary Computation is especially suitable for
those problems in which a cost–effort trade–off exists
such as event correlation (Suarez-Tangil et al., 2009).
2 POSITION STATEMENT
As very promising solutions which are emerging by
some sort of biological inspiration, Artificial Immune
Systems
1
(AISs) have been proven to contribute im-
portant benefits to different areas within computer se-
curity, since efficient abstractions of processes were
found in the mid 1980s by (Farmer et al., 1986). In
particular, several works focus on analyzing how im-
munological concepts may be applied to intrusion de-
tection (Kim et al., 2007), pattern recognition and
1
The immune network theory was first introduced by Jerne
(Jerne, 1974) as a way to explain the memory and learning capa-
bilities exhibited by the immune system. This theory has inspired a
subfield of optimization algorithms as many other fields unrelated
to biological immunology.
422
Suarez-Tangil G., Palomar E., Pastrana S. and Ribagorda A..
ARTIFICIAL IMMUNITY-BASED CORRELATION SYSTEM.
DOI: 10.5220/0003610604220425
In Proceedings of the International Conference on Security and Cryptography (SECRYPT-2011), pages 422-425
ISBN: 978-989-8425-71-3
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
classification (Carter, 2000), to name a few. Perhaps
the main advantage of AISs is that not only super-
vised learning is possible (Watkins et al., 2004), but
also unsupervised (De Castro and Timmis, 2002) in-
deed. In this position paper, we extend the typical
architecture of a SIEM system to efficiently introduce
an adaptive learning framework upon the correlation
process, considering the following statements:
Generally, the existing SIEM tools present limita-
tions and contextual constrains. In addition, cur-
rent SIEM frameworks deploy their own architec-
ture. We propose a global framework which inte-
grates the most promising research advances and
formalizes a unified architecture design towards
an intelligent correlation system.
The strategy of combining intelligence and self–
adaptation to optimize different types of comput-
ing services is emerging as a robust and efficient
approach. Therefore, AIS-based SIEM systems
will facilitate an adaptive correlation of novel
multi–step attacks.
Authors in (Hofmeyr and Forrest, 2000) propose
LISYS, a framework for applying AIS to network
security. Our approach presents two main differ-
ences. On the one hand, LISYS applies an AIS
architecture to Network Intrusion Detection Sys-
tem (NIDS) whereas our application domain con-
siders a centralized SIEM. Specifically, our ap-
proach could receive inputs from LISYS as a sen-
sor to correlate complex multi-step attacks. On
the other hand, our approach adds honeynets in or-
der to minimize human intervention as far as pos-
sible. To the best of our knowledge, the AIS has
not been relevant to the context of security event
correlation by applying SIEM.
The rest of the paper is organized as follows.
Section 3 describes the foundations of our work–in–
progress, establishing the AIS entities needed to in-
troduce an adaptive learning component into a tradi-
tional event correlation engine. Finally, in Section 4
we establish the main conclusions as well as the im-
mediate future work.
3 AN ARTIFICIALLY IMMUNE
SIEM ARCHITECTURE
We envision the construction of a SIEM system using
an artificial immune network model as a three-layer
architecture (Fig. 1) comprising the following build-
ing blocks:
The physical barrier offers a physical protection
PHYSICAL
BARRIER
(IPS)
INNATE
IMMUNE
SYSTEM
(IDS)
ADAPTIVE IMMUNE SYSTEM (ILS)
IMMUNOLOGICAL MEMORY
PATTERN
RECOGNITION
RECEPTOR
Internet
DMZ Traffic
Darknet
Traffic
Patterns
Incoming
Traffic
Figure 1: An event correlation framework based on AIS.
against pathogens attacking the system as some
sort of prevention layer.
The Innate Immune System (present at birth in hu-
mans) deploysimmune agents which are in charge
of protecting the system against invaders as well
as providing pattern recognition mechanisms.
The Adaptive Immune System defines the logic
to learn, adapt and memorize antigens and secrete
the appropriate anti-body (i.e. correlation rules).
The complex adaptive system is the main focus of
interest here that involves diverse and multiple inter-
connected elements, which tend to provide capacity
to change and learn from experience. Several works
have partially applied AIS to specific domains. An in-
teresting approach for mapping the immunity entities
and process on to the development of computational
models is presented in (Dasgupta, 2006). In the fol-
lowing sections we further elaborate on the essential
concepts which leads to an AIS-based implementa-
tion.
3.1 Application Domain
We should consider the event correlation process from
two different viewpoints:
Definition 3.1 (Automatic Supervision). The process
to extract the knowledge related to a novel attack
without human intervention. Each extraction will
form temporal correlation rules.
Definition 3.2 (Expert Supervision). An expert
guides the process to extract knowledge related to a
novel attack and forms a permanent correlation rule.
For the sake of completeness, we include the fol-
lowing basic concepts:
Definition 3.3 (Event Fingerprint). Set of attributes
which identifies certain event’s properties. The speci-
fication of these attrib. depends on the SIEM system.
Event correlation aims at obtaining the fingerprint
of a series of aggregated events. Thus,
ARTIFICIAL IMMUNITY-BASED CORRELATION SYSTEM
423
Definition 3.4 (Event Aggregation). Event aggrega-
tion gathers together a collection of events which ful-
fill particular premises.
Definition 3.5 (Event Correlation). Probabilistically
define the relationship between a set of aggregated
events.
And consequently,
Definition 3.6 (Correlation Fingerprint). Represents
a correlation produced as a response of the successful
relationship between a set of attributes.
We position that intruder’s actions swiftly evolve
to become more effective, as well as more sophisti-
cated generations of malware, i.e. polymorphic mal-
ware. In this regard, malware-analysis tools inte-
grated along with an adaptive learning system will in-
tegrate our architecture to automatically generate spe-
cific correlation-fingerprints’.
Definition 3.7 (Darknet). Also known as network
telescope, is a system used to observe different large-
scale events by monitoring unused network addresses.
Therefore, we can generate fake interactions with
the intruder by emulating common exposed services.
We assume here that despite aforementioned intrusion
evasions, users habits and used services will tend to
behavealike. Thus, evolvedmalware will still interact
with our emulated services. Honeynet projects can be
used in this regard (Spitzner, 2003).
3.2 Representation
A key principle within an AIS is now introduced,
namely the proteins.
Definition 3.8 (Proteins). Artificial immune theory
defines the concept of secreting proteins as the mech-
anism used to detect non–self pathogens –malicious
cells– which in turn are destroyed by antibodies. Pro-
teins constitute the parameters to monitor and then
distinguish self and non-self behaviors.
Now, two approaches for representing the appli-
cation domain are possible: (i) when nodes represent
the proteins role, and/or (ii) when events act as the
subject to monitor. The former focuses on learning
about the nodes which exhibit anomalousbehavior. In
this context, honest and misbehaving nodes embody
self and non-self cells respectively. A major draw-
back of this approach is that compromised nodes gen-
erally produce both types of traffic, and therefore this
could cause serious problems of false classification.
However, the alternative seems promising as identify-
ing the events related to a certain attack present more
similarities with traditional SIEM procedures.
We formulate the bases of the artificial immune
representation (see Fig. 2) as follows. The examined
Figure 2: Mapping between immune and correlation enti-
ties.
network depicts the biological body and the events
produced on the AIS will be represented as proteins.
Events can be classified as either self or non-self
events. Thus, authorized activity will be classified
as self by sensors and the opposite is cataloged as
non–self. Additionally, on the one hand, an antigen
is a series of events that matches an event fingerprint.
And on the other hand the antibody is the pattern (or
rule) responsible of identifying a specific sequence of
events. Therefore,
Claim 3.1 (Proteins). Proteins are the network ac-
tivity monitored by sensors connecting to a central
SIEM. Proteins identify correlation fingerprints.
3.3 Immune Algorithm
In this section, we position the artificial immune algo-
rithm as a three-phase protocol which combines tradi-
tional IDS concepts, data mining, honeynets and, just
when strictly needed, the expert supervision, as fol-
lows:
I Initial Innate Immune Definition. In this
phase, the expert must define a range of val-
ues for the collection of correlation fingerprints.
These values will act as discriminators for self
cells. How to implement the appropriate value
on the corresponding correlation fingerprint is
SECRYPT 2011 - International Conference on Security and Cryptography
424
critical. To this regard, existing repositories for
known attacks and their associated correlation
rules may be useful. Generally, this repository
is named Gene Library and is essential for the
process below.
II Adaptive Algorithm. The adaptive algorithm
produces a number of correlation fingerprints
that will be used to learn new correlation rules.
The basis of this algorithm relies on the AIS
principles –random adaptations will produce new
patterns by applying (i) antibody secretion, (ii)
negative selection, (iii) pathogen match, and (iv)
clonal selection based on affinity mutation.
III Adaptive Immunological Memory Consolida-
tion. New correlation fingerprints are consoli-
dated based on the following knowledge extrac-
tion: using the expertise of the administrator and
automation techniques. On the one hand, the
expert has to manually inspect and validate the
correlation rules in terms of their accuracy. On
the other hand, honeynets seem the best candi-
date to assist the automated consolidation pro-
cess. Specifically, generated fingerprints will be
validated using the non-self activity reported on
the darknet at the beginning. If any of the finger-
prints matches then the immunological memory
(associated to each correlation) will be increased.
4 CONCLUSIONS
In this position paper, we have discussed the appli-
cation of AIS techniques to optimize current SIEM
systems. To this regard, we propose an adaptive im-
mune correlation system to be included into a typi-
cal SIEM architecture. Our global objective is to effi-
ciently generate correlation rules and adaptively pre-
dict novel multi-step attacks. Our proposal comprises
various strategies already used in intrusion detection,
data mining, honeynetsand, just when strictly needed,
the expert supervision. Our hope is that this position
paper will, directly or indirectly, inspire new direc-
tions on applying intelligence to security event corre-
lation.
REFERENCES
Anuar, N., Papadaki, M., Furnell, S., and Clarke, N. (2010).
An investigation and survey of response options for
Intrusion Response Systems. In Information Security
for South Africa (ISSA), 2010, pages 1–8. IEEE.
Carter, J. H. (2000). The immune system as a model for
pattern recognition and classification. Journal of the
American Medical Informatics Association: JAMIA,
7(1):28–41.
Dasgupta, D. (2006). Advances in ais. IEEE Comp. Intelli-
gent Magazine, 1(4):40–49.
De Castro, L. and Timmis, J. (2002). Artificial immune
systems: a new computational intelligence approach.
Springer Verlag.
Farmer, J. D., Packard, N. H., and Perelson, A. S. (1986).
The immune system, adaptation, and machine learn-
ing. Physica D: Nonlinear Phen., 22(1-3):187–204.
Hofmeyr, S. A. and Forrest, S. (2000). Architecture for an
artificial immune system. Evolutionary computation,
8(4):443–73.
Jerne, N. K. (1974). Towards a network theory of the im-
mune system. Ann. Immunol., 125C:373–389.
Kim, J., Bentley, P., Aickelin, U., Greensmith, J., Tedesco,
G., and Twycross, J. (2007). Immune system ap-
proaches to intrusion detection–a review. Natural
computing, 6(4):413–466.
Lippmann, R. P. and Cunningham, R. K. (2000). Improv-
ing intrusion detection performance using keyword
selection and neural networks. Computer Networks,
34(4):597–603. Recent Advances in IDS.
Mutz, D., Vigna, G., and Kemmerer, R. (2003). An Expe-
rience Developing an IDS Stimulator for the Black-
Box Testing of Network Intrusion Detection Systems.
In Proc. of the 2003 Computer Security Applications
Conf., Las Vegas, Nevada.
Ripley, B. (1994). Neural networks and related methods for
classification. Journal of the Royal Statistical Society,
56(3):409–456.
Spitzner, L. (2003). The honeynet project: Trapping the
hackers. IEEE Security and Privacy, pages 15–23.
Suarez-Tangil, G., Palomar, E., Fuentes, J. D., Blasco, J.,
and Ribagorda, A. (2009). Automatic rule generation
based on genetic programming for event correlation.
In Computational Intelligence in Security for Infor-
mation, Advances in Soft Computing, pages 127–134,
Burgos, Spain. Heidelberg, Springer Berlin.
Watkins, A., Timmis, J., and Boggess, L. (2004). Artificial
Immune Recognition System: An Immune-Inspired
Supervised Learning Algorithm. Genetic Program-
ming and Evolvable Machines, 5(3):291–317.
ARTIFICIAL IMMUNITY-BASED CORRELATION SYSTEM
425