Sei sulla pagina 1di 6

Enhancing effectiveness of intrusion detection

systems: A hybrid approach

Basant Subba , Santosh Biswas, Sushanta Karmakar


Department of Computer Science & Engineering
Indian Institute of Technology, Guwahati
Assam, India 781039
Email: s.basant@iitg.ernet.in, santosh biswas@iitg.ernet.in , sushantak@iitg.ernet.in

AbstractIntrusion Detection Systems (IDSs) proposed in the packets on the network and compares them against a database
literature can broadly be classified as either signature based or of well known attack signatures. The attack signatures are
anomaly based. Although both these classes of IDSs effectively combination of the packet header feature values and packet
detect wide range of network attacks, they have their own set content descriptions intended for identification of anomalous
of drawbacks. Signature based IDSs are incapable of detecting data traffic. The main advantage of signature based IDSs is
new attacks and produce a large number of false positive alarms
when operated with default settings. On the other hand, anomaly
that they have fairly high accuracy and detection rate against
based IDSs require extensive training before deployment and known attacks. However, they fail to detect new attacks whose
are computationally expensive. In this paper, we aim to address signatures has not yet been updated in the databases. In
these issues by proposing an efficient hybrid intrusion detection addition, most of the commercially available signature based
framework with high detection rate and low false alarm rate. IDSs like Snort [2], Bro [3] etc. uses regular expressions
A novel false alarm minimization technique is used to reduce instead of using an exact matching string when specifying
the false alarm rate of the signature based component and attack signatures. This results in a large number of false alarm
a simple header based anomaly detection module is used to generation when they are deployed with default settings.
minimize the computational overhead of the anomaly based
component. Experimental results on the benchmark DARPA On the other hand, anomaly based IDSs consist of two
IDEVAL dataset and an in-house test bed dataset show that the distinct phases namely, the training phase and the testing
proposed framework achieves a high detection rate and accuracy phase. During the training phase the normal profile of the
across a wide range of network attacks, while at the same time
network is learned by the anomaly based IDSs and during
minimizes the overall computational overhead.
the testing phase the learned model of normal profile is
KeywordsIntrusion Detection Systems (IDSs), DARPA IDE- applied to the current network traffic to detect intrusions.
VAL dataset, False alarms, Snort, PHAD Depending on the methods employed, anomaly based IDSs
can be categorized into statistics based [5], machine learning
I. I NTRODUCTION based [6] [7] and data-mining based [8] [9] [10] IDSs. The
Rapid proliferation of heterogeneous networking devices main advantage of anomaly based IDSs is that unlike the
that can seamlessly communicate and interact with each other signature based IDSs they do not rely on any predefined attack
has significantly increased the complexity of the network secu- signatures to detect network intrusions and are therefore able to
rity. Most of these devices have inherent design flaws that leads detect novel attacks. However, the main drawback of anomaly
to various security threats. Various preventive mechanisms based IDSs is the difficulty involved in defining the normal
like cryptography, authentication, data encryption, firewall etc., network profile against which the traffic being monitored are
have been proposed in the literature as a first line of defense correlated to detect network intrusions. Estimating and defining
to prevent network intrusions. However, these mechanisms the normal network profile, especially in a dynamic network,
have their limitations and attackers are increasingly finding where the traffic pattern varies at regular interval is a difficult
their way around them to gain unauthorized access to the task. Another key issue of anomaly based IDSs is their high
network and launch attacks. Moreover, these mechanisms computational overhead. Most of the anomaly based IDSs
cannot detect inside attacks, where the attacker is a legitimate employ a deep level packet inspection which increases their
part of the network. To address these security issues, Intrusion computational overhead and limits their performance in a high
Detection System (IDS) has been proposed in the literature speed networks.
as a complementary second line of defense[1] [2] [3] [4].
In this paper, we aim to address these issues of signature
IDS monitors network for sign of intrusions and raises an
and anomaly based IDSs by proposing a new hybrid IDS
alarm whenever any malicious or intrusive network activities
framework. The proposed hybrid framework combines the
are detected. Therefore, IDS helps to contain the network
desirable properties of both the signature and anomaly based
breach by taking appropriate preventive measure (raising alarm
IDSs, while eliminating the drawbacks associated with them.
and reporting to network administrator) before any significant
The signature based component of the proposed framework
damage can be inflicted on the network by the intruder.
uses a novel false alarm minimization technique to minimize
Based on their mode of detection, IDSs can broadly be the overall false positive alarm rate while the anomaly based
classified into following two categories namely, signature component uses a simple packet header anomaly detector
based and anomaly based. Signature based IDSs monitor data (PHAD) [4] module to minimize the overall computational
overhead. PHAD inspects only the packets header field with- leads to an accurate anomaly detection model. Their scheme
out performing any in-depth payload inspection. In addition,the was able to detect wide range of anomalies with fairly high
proposed framework is capable of detecting network attacks sensitivity and accuracy. Kim et al. [16] proposed IDS scheme
that the signature and anomaly based IDSs cannot detect alone that uses address correlation properties to detect anomalies.
in isolation. Experimental results on the benchmark DARPA Their scheme examines the packet headers rather than the
IDEVAL dataset and the testbed dataset show that the proposed packet payload, to look for the correlation between various
framework achieves a high detection rate and accuracy against header fields using the wavelet analysis. The drawbacks of
wide range of attacks while at the same time minimizes the these IDS schemes are their high computational overhead.
overall computational overhead.
Yen et al. proposed a hybrid IDS framework for wireless
The rest of the paper has been structured in following ways. sensor networks [17]. Their framework consists of a mis-
Section II discusses the related works on false alarm mini- use detection model, back propagation network (BPN) based
mization and computational overhead optimization of IDSs. anomaly detection model and decision making modules. In
Section III presents the overall description of the proposed their proposed framework, the data packets are first analyzed
hybrid intrusion detection framework. Section IV presents the by the anomaly detection model. If any anomalies are detected
experimental results and finally we conclude with conclusion in the data packets then they are forwarded to both the misuse
and future works in section V. based model and the decision making model. The misuse based
model analyzes the data packets further for sign of intrusions.
II. R ELATED W ORK Finally, the decision making module combines the outputs of
both models to determine whether or not the data packets are
IDS frameworks that correlate alarms generated by the anomalous. In case of intrusion, the decision making module
signature based IDS with network vulnerabilities based on reports to the base station. However, the drawback of this
their reference numbers are proposed in [11] [12]. In these scheme is its high computational overhead.
frameworks, various vulnerability scanners are employed to
develop the Vulnerability Threat Profile of the network. IDS Summarizing the related works, we found that the false
alarms are then correlated with vulnerabilities in the Vulner- alarm minimization techniques proposed in the literature de-
ability Threat Profile based on their reference numbers. If creases the detection rate of the signature based IDSs. We
the IDS alarm cannot be correlated with any of the network also found that most of the anomaly based IDSs proposed
vulnerabilities then it is treated as false positive alarm and in the literature are computation intensive due to their deep
ignored. However, the drawback of these schemes is that level packet inspections which puts a severe limitation on their
some of the IDS alarms may not have an associated reference performance in a high speed network. The main contribution
numbers and therefore, they cannot be correlated with any of of this paper are as follows:
the network vulnerabilities. This increases the false negative
rate and hence lowers the detection rate of the IDS. 1) Propose a hybrid intrusion detection framework with
high accuracy and detection rate.
Valdes et al. [13] applied Bayesian networks to detect 2) Reduce the overall false alarm rate of the IDS by
anomalies on burst of traffic. Their Bayesian network assigns proposing a novel false alarm minimization scheme.
a probabilistic relationship to various variables of the data traf- 3) Minimize the computational overhead of IDS by
fic and is capable of determining interdependencies between analyzing only the data packets header field using
various variable of the data traffic. They showed that their a Packet Header Anomaly Detector (PHAD).
scheme is capable of detecting the DDoS attack that will not be
detected if each of the attack components were to be examined
III. P ROPOSED H YBRID I NTRUSION D ETECTION
separately. Dickerson et al. [9] proposed a simple data mining
FRAMEWORK
technique that processes the network data traffic and generates
a set of fuzzy rules for every feature under observation which In this section, we provide a detailed description of the
detects individual attacks based on each of the features. The proposed hybrid intrusion detection framework. The proposed
main drawback of theses schemes are their high computational framework comprises a signature based and an anomaly based
overhead. components. Various enhancement mechanisms are employed
Barford et al. [14] proposed a statistical anomaly based IDS to fine tune the performance of both these components to
which operates under the assumption that the characteristics minimize the overall false alarm rate and the computational
of the anomalous network traffic significantly deviates from overhead of the proposed framework. A detailed descriptions
the normal pattern, in terms of the number of data bytes about each of these components are provided in subsequent
exchanged, packets transmitted, heavy usage of certain set of sections.
IP addresses and ports etc. Their scheme was successful in
identifying large traffic changes such as bandwidth flooding A. Signature based component
attacks. However, the drawback of their scheme is that it
fails to detect attacks if the attacker is smart enough to keep Most of the signature based IDSs use regular expressions
the disruptions caused by the attacks below certain threshold instead of an exact matching string when specifying attack
levels. signatures. Therefore, they produce a large number of false
positive (FP) alarms when operated with default settings. To
IDS scheme that uses entropy as a tool to summarize vari- address this issue, we propose a novel false alarm minimization
ous traffic features was proposed by Lakhina et al. [15]. They technique to reduce the overall false alarm rate of the signature
showed that the network traffic feature distributions analysis based IDS. The overall architecture of the signature based
Fig. 1: Architecture of the signature based component

TABLE I: Snapshot of alarms generated by signature based IDS


Src IP Addr Src port Dest IP Addr Dest port Ref No. Protocol Content OS
EXTERNAL NET 80 172.16.112.0/24 4502 CVE-1999-0874 TCP cmd.exe Windows
... ... ... ... ... ... ... ...
EXTERNAL NET 22 172.16.112.0/24 90 CVE-2001-0731 TCP s|00|p|00| |00|s|00| Solaris
EXTERNAL NET 25 172.16.112.0/24 1503 IP Microsoft Windows Windows
EXTERNAL NET 23 172.16.112.0/24 110, 25 CVE-1999-0146 IP Microsoft Windows Windows

TABLE II: Snapshot of vulnerabilities in the Vulnerability Threat Profile


Src IP Addr Src port Dest IP Addr Dest port Ref No. Protocol Content OS
EXTERNAL NET any HOME NET 4502, 25, 80 CVE-1999-0874 TCP cmd.exe Windows
... ... ... ... ... ... ... ...
EXTERNAL NET any HOME NET 3302, 403, 80 CVE-2003-0316 TCP | 5C|../|00 00 00| Linux
EXTERNAL NET any HOME NET 90,22, 25 CVE-2001-0731 TCP s|00|p|00| |00|s|00| Solaris
EXTERNAL NET any HOME NET 1402, 80, 22 CVE-2002-0012 TCP /bin Redhat
EXTERNAL NET any HOME NET 25,80 NessusID: 10280 TCP /root SunOS

component of the proposed framework is shown in Fig. 1. A Vulnerability Threat Profile: The network vulner-
brief description about its various modules are provided below: abilities detected by the vulnerability scanners are
augmented with various network context information
Packet sniffer: This module captures the incoming parameters like reference numbers, IP addresses, port
data packets for intrusion detection analysis. It uses numbers, protocols, traffic types etc., to create the
the Libpcap and Wincap libraries for capturing data Vulnerability Threat Profile of the network. Alarms
packets. raised by the signature based IDS are correlated with
Signature Based IDS: This module correlates the vulnerabilities in the Vulnerability Threat Profile. Ta-
data packets captured by the packet sniffer module ble II shows the snapshot of the vulnerabilities in the
with attack signatures in the signature database. If Vulnerability Threat Profile.
the captured packet matches with one or more attack
signatures in the database, an alarm is raised by the Binary Correlation Vector (BCV): Signature based
IDS. Snort [2] is used as a default signature based IDS IDS alarms are correlated with vulnerabilities in the
in our study. Vulnerability Threat Profile to obtain the BCV. Let
the ith alarm raised by the signature based IDS be
IDS Alarms: It is output of the signature based represented by an alarm vector Ai = ha1 , a2 , ...., an i.
IDS and contains both the relevant and non-relevant Similarly, let the jth vulnerability in the Vulnerability
alarms. Table I shows the snapshot of alarms generated Threat Profile be represented by the vulnerability
by the signature Based IDS. vector V j = hv1 , v2 , ...., vn i. The positional match or
Vulnerability Database: This module contains attack mismatch between Ai and V j is represented by the
signatures populated from various sources such as BCV of the form C = hc1 , c2 , ...., cn i, where ci = 0,
CVE database [18], Bugtraq [19] etc. if (ai == vi ), else it is set to 1.
Vulnerability Scanner: This module scans the net- Global Vector Table (GVT): This module consists
work to identify various vulnerabilities present in the of set of BCVs corresponding to valid attacks on the
network. Different vulnerability scanners like Nessus network. Each BCV in the GVT of Table III is a binary
[20], Nmap [21], Rapid7 Nexpose etc., are used to vector corresponding to the following tuple hDest IP,
obtain an exhaustive list of all the network vulnera- Dest port, Src IP, Src port, Ref No, Protocol, Content,
bilities. OSi. If the BCV generated after correlating the IDS
TABLE III: Global Vector Table with a list of BCVs or more vulnerabilities in the Vulnerability Threat Profile are
corresponding to valid attacks treated as True Positives (TPs) while others are discarded as
False Positives (FPs). On the other hand, the anomaly based
Binary Correlation Vector Description component analyzes the header fields of the data packets for
h 0, 0, 0, 0, 0, 0, 0, 0 i All feature matches sign of anomalies. Data packets marked as anomalous by the
h 0, 1, 0, 0, 0, 0, 0, 0 i Mismatch in Dest Port No. anomaly based component (PHAD) are treated as true alarms.
h 0, 0, 1, 0, 0, 0, 0, 0 i Mismatch in Src IP address
h 0, 0, 0, 1, 0, 0, 0, 0 i Mismatch in Src Port No.
h 0, 0, 0, 0, 1, 0, 0, 0 i Mismatch in Reference No. IV. E XPERIMENTAL R ESULTS
h 0, 0, 0, 0, 0, 1, 0, 0 i Mismatch in Protocol
h 0, 0, 0, 0, 0, 0, 1, 0 i Mismatch in Content The performance of the proposed hybrid intrusion detection
framework was tested on the benchmark DARPA IDEVAL
dataset [22] and an in house test bed dataset. Accuracy and
alarm with vulnerabilities in the Vulnerability Threat detection rate were used as parameters for the evaluation of
Profile has a corresponding entry in the GVT, then the proposed framework.
the alarm corresponding to that BCV is marked as
true positive (TP) alarm, else it is ignored as a false A. Performance analysis on the test bed dataset
positive alarm. In general, IDS alarms are flagged as
a TPs if they satisfy the following conditions: To create the testbed dataset, a heterogeneous test bed
network comprising of several hosts with various operating
1) The source IP address of the IDS alarm and
systems like Windows 2000 Server, Windows XP, Windows
the vulnerability being correlated matches.
7, Ubuntu 12.04, Redhat 7.2 and Fedora 12 running various
2) The OS type of the IDS alarm and the vulner-
applications such as Telnet, SSH, FTP server, SQL server
ability being correlated matches.
etc. was set up. Nmap [12] and Nessus [8] vulnerability
3) There is at most one mismatch in any other
scanners were used to scan the test bed network and build
features between the IDS alarm and the vul-
the Vulnerability Threat Profile. Some of the machines in the
nerability being correlated.
test bed was used to generate attacks using the metasploit [23],
which is a freely available open source exploit toolbox. The
B. Anomaly based component types of attack considered in the test bed setup were Denial of
The anomaly based component of the proposed hybrid IDS Service (DoS) attack, probing attack, remote code execution,
framework uses the Packet Header Anomaly Detector (PHAD) SQL injection, Microsoft server exploits etc.
[4] to detect anomalous data packets. PHAD inspects only the Wireshark [24] was used as packet sniffer to record the
packets header field without performing any in-depth payload data traffic information of the testbed network (as pcap files).
inspection. It learns the normal range of values of different Two different set of datasets were generated from the given
fields of the packets header at the transport (TCP, UDP), testbed network. The first dataset comprising of only the
network (IP) and data link (Ethernet) layers during its training normal network traffic without any attacks was used to train
phase. It operates under the assumption that rarer events are the PHAD. The second dataset, which consists of both the
more likely to be intrusive and hostile. If an event occurs with a normal data traffic and attacks was used as a test dataset for
probability q, PHAD assigns it a score of 1/q. PHAD assumes both the signature and the anomaly based components.
that if an event last occurred t seconds ago, then the probability
that it will occur in the next one second is given by 1/t. A total of 48 attack instances are present in the test dataset
Each packet header feature containing an anomalous value is of the test bed network. Table IV shows the performance
assigned an anomalous score given by score f eat = tn/r, where n analysis of the proposed hybrid IDS framework on the test
is the count of number of times the given features are observed bed dataset. As shown in the table, 37 attack instances were
during the training phase and r is the number of distinct values detected by the anomaly based component while producing 13
observed for this feature during the training phase. Finally, false positive alarms. On the other hand, a total of 93 alarms
the packet is scored by summing up the individual anomaly were generated by the signature based component initially.
scores of the features present in the packet, i.e. score packet = However, after application of the false alarm minimization
ti ni /ri , where anomalous features are the features technique the number of alarms was reduced to 35. Out of
ianomaly f eat these 35 alarms, 32 were true positives while 3 of them
with values not observed during the training phase. were false positives. In general, a total of 43 attack instances
were detected by the proposed hybrid IDS framework with 27
C. Overall architecture of the proposed framework common attacks that were detected by both the signature and
anomaly based components. The accuracy (ACC) and detection
Fig. 2 shows the overall architecture of the proposed hybrid rate (DR) of the proposed hybrid IDS framework on the test
intrusion detection framework. As shown in the figure, the bed dataset was found to be 72.88% and 89.58%, respectively.
incoming data packets are analyzed by both the signature
based component (Snort) and the anomaly based component
B. Performance analysis on the DARPA IDEVAL dataset
(PHAD) simultaneously. If the data packets being analyzed
by the signature based component matches with one or more To analyze the performance of the proposed hybrid IDS
attack signatures in the database then an alarm is raised. These framework on the benchmark DARPA IDEVAL dataset, the
alarms are then re-evaluated using the false alarm minimization anomaly based component (PHAD) was trained on 5 days of
technique discussed earlier to filter out the false positive attack free network traffic of the third week dataset. The trained
alarms. Only those alarms which can be correlated with one model was then tested on 9 days of dataset from week 4 and
Fig. 2: Oveall architecture of the proposed hybrid IDS framework

TABLE IV: Performance of the proposed hybrid IDS TABLE VI: Performance analysis of Snort on 4th and 5th
framework on test-bed dataset week of IDEVAL Test dataset
Anomaly Based Signature Based Performance Day No of alarms Alarms after Corr. TP FP
Component Component Analysis
No. of No. of Alarms after Mon4 87 8 6 2
TP FP TP FP ACC DR Wed4 107 9 5 4
Alarms Alarms correlation
Thu4 91 13 7 6
50 37 13 93 35 32 3 72.88% 89.58%
Fri4 109 11 8 3
Mon5 116 12 7 5
Tue5 97 10 6 4
Wed5 83 15 9 6
week 5 during which there were 201 instances of about 56 Thu5 113 7 4 3
types of attacks. Table V lists various attacks detected and Fri5 102 14 5 9
the false positive alarms generated by PHAD based on the
anomaly values observed in the given packet header field. Out
of 201 attack instances, 65 attacks were detected by PHAD TABLE VII: Performance analysis of different IDS
while generating a total of 102 false positive alarms. frameworks on the IDEVAL Test dataset
IDS scheme TP alarms FP alarms ACC DR
TABLE V: Performance analysis of PHAD on 4th and 5th Snort 57 813 6.55% 28.36%
PHAD 65 102 38.92% 32.34%
week of IDEVAL dataset HYB IDS 113 144 43.96% 56.22%

Field TP FP
Ethernet Size 1 1
Ethernet Dest Hi 1 5 PHAD and the proposed hybrid intrusion detection framework
Ethernet Src Hi 0 7
IP TOS 0 6 (HYB IDS) on the 9 days of IDEVAL test dataset. It can be
IP Packet Length 2 3 observed from the Table that the performance of Snort on the
IP TTL 31 19 DARPA IDEVAL dataset was very poor. One of the reason
IP Frag Ptr 7 4
IP Source Address 4 1 for this poor performance could be the fact that attacks of
IP Dest Address 2 9 the IDEVAL dataset are obsolete and Snort no longer includes
TCP Acknowledgment 0 0
TCP Flags UAPRSF 7 3
rules to detect them. In addition, Snort also generated a large
TCP Window Size 0 1 number of FP alarms (813). It can be observed from the
TCP Checksum 1 28 Table that the proposed HYB IDS framework has the highest
TCP URG Ptr 3 5
TCP Options 2 3 accuracy (ACC) and detection rate (DR) among the three
UDP Length 0 6 schemes.
UDP Checksum 2 0
ICMP Checksum 2 1 Summarizing the result, we found that the number of
attacks detected by the Snort is comparatively low (57), while
Table VI shows the list of attacks detected and the false it generates a significantly large number of false positive
alarms generated by Snort on the IDEVAL test dataset be- alarms (813). The performance of PHAD was better than Snort.
fore and after the application of the proposed false alarm It detected 65 attacks and generated only 102 false positive
minimization technique. The Vulnerability Threat Profile of alarms. Finally, the proposed HYB IDS framework detected
the IDEVAL network was developed by running Nessus and 113 attacks while generating a total of 144 FP alarms (42
Nmap vulnerability scanners on various operating systems from Snort and 102 from PHAD). The high detection rate of
(Redhat, Solaris, SunOS, Windows NT, Windows 95 etc.) of the HYB IDS framework can be attributed to the fact that the
the IDEVAL network. It can be observed from the table that type of attacks detected by the Snort and PHAD are relatively
the proposed false alarm minimization scheme significantly different with a limited number of common attacks detected
reduces the number of alarms generated by the Snort. between them. Similarly, the proposed HYB IDS framework
achieves the highest accuracy among all the schemes since
Table VII shows the performance analysis of Snort, most of the false positive alarms generated by its signature
TABLE VIII: Performance comparison of various intrusion detection frameworks on the DARPA IDEVAL dataset
IDS models Snort [2] PHAD [4] HYB IDS Pure anomaly based IDS [8] [7]
Detection rate Very Low Low High Very High

False alarm rate Very High Low Low Very Low

Computational overhead Low Low Low Very High

High detection rate Detects both known and unknown Low computational overhead Very high detection rate
Advantage
against known attacks attacks with low computational overhead with high detection rate and very low false alarm rate

Very high false alarm rate,


Disadvantage Limited detection capability Moderate false alarm rate Very high computational overhead
cannot detect new attacks

based component (Snort) are filtered out by the false alarm [8] Y. Yu and H. Wu, Anomaly intrusion detection based upon data
minimization module. mining techniques and fuzzy logic, in IEEE International Conference
on Systems, Man, and Cybernetics, Oct 2012, pp. 514517.
Performance comparison of the proposed HYB IDS frame- [9] J. Dickerson and J. Dickerson, Fuzzy network profiling for intrusion
work with other intrusion detection frameworks on the DARPA detection, in 19th International Conference of the North American
IDEVAL dataset is shown in tabular form in Table VIII. Fuzzy Information Processing Society, 2000, pp. 301306.
[10] S. M. Bridges and R. B. Vaughn, Fuzzy Data Mining And Genetic
Algorithms Applied To Intrusion Detection, in Proceedings of the
V. C ONCLUSION AND F UTURE W ORK National Information Systems Security Conference (NISSC, 2000, pp.
1619.
Both signature and anomaly based IDSs have their own set [11] F. Massicotte, M. Couture, L. Briand, and Y. Labiche, Model-Driven,
of drawbacks. Signature based IDSs are incapable of detecting Network-Context Sensitive Intrusion Detection, in Model Driven En-
new attacks and produce a large number of false positive gineering Languages and Systems, 2007, vol. 4735, pp. 6175.
alarms when operated with default settings, while anomaly [12] S. Neelakantan and S. Rao, A Threat-Aware Anomaly-Based Intrusion-
based IDSs are computationally intensive. To address these Detection Approach for Obtaining Network-Specific Useful Alarms,
in Proceedings of the 10th International Conference on Distributed
issues, we proposed a new hybrid intrusion detection frame- Computing and Networking, 2009, pp. 175180.
work with high level of performance and low computational [13] A. Valdes and K. Skinner, Adaptive, Model-Based Monitoring for
overhead. The proposed hybrid intrusion detection framework Cyber Attack Detection, in Recent Advances in Intrusion Detection,
uses a novel false alarm minimization technique to reduce the ser. Lecture Notes in Computer Science, 2000, vol. 1907, pp. 8093.
overall false alarm rate of its signature based component and [14] P. Barford, J. Kline, D. Plonka, and A. Ron, A Signal Analysis of Net-
a simple header based anomaly detection module to minimize work Traffic Anomalies, in Proceedings of the 2nd ACM SIGCOMM
the computational overhead of its anomaly based component. Workshop on Internet Measurment, ser. IMW 02, 2002, pp. 7182.
For our future work, we aim to further enhance the detection [15] A. Lakhina, M. Crovella, and C. Diot, Mining Anomalies Using Traffic
Feature Distributions, SIGCOMM Comput. Commun. Rev., vol. 35,
rate and accuracy of both the anomaly and the signature no. 4, pp. 217228, Aug. 2005.
based components of the proposed framework through various
[16] S. Kim, A. Reddy, and M. Vannucci, Detecting Traffic Anomalies
refinement processes. through Aggregate Analysis of Packet Header Data, in Networking
2004, ser. Lecture Notes in Computer Science, 2004, vol. 3042, pp.
10471059.
R EFERENCES
[17] K. Q. Yan, S. C. Wang, S. S. Wang, and C. W. Liu, Hybrid Intrusion
[1] A. Abduvaliyev, A.-S. Pathan, J. Zhou, R. Roman, and W.-C. Wong, Detection System for enhancing the security of a cluster-based Wireless
On the Vital Areas of Intrusion Detection Systems in Wireless Sensor Sensor Network, in 3rd IEEE International Conference on Computer
Networks, IEEE Communications Surveys Tutorials, vol. 15, no. 3, pp. Science and Information Technology (ICCSIT), vol. 1, July 2010, pp.
12231237, 2013. 114118.
[2] M. Roesch, Snort - Lightweight Intrusion Detection for Networks, in [18] CVE-common vulnerabilities and exposures, 2013.
Proceedings of the 13th USENIX Conference on System Administration, [19] F. Massicotte, M. Couture, and Y. Labiche, Context-Based Intrusion
1999, pp. 229238. Detection Using Snort, Nessus and Bugtraq Databases. in PST, 2005.
[3] V. Paxson, Bro: A System for Detecting Network Intruders in Real- [20] J. Beale, R. Deraison, H. Meer, R. Temmingh, and C. V. D. Walt, Nessus
time, Computer Networks, vol. 31, no. 23-24, pp. 24352463, Dec. Network Auditing. Syngress Publishing, 2004.
1999.
[21] G. F. Lyon, Nmap Network Scanning: The Official Nmap Project Guide
[4] M. V. Mahoney and P. K. Chan, PHAD: Packet Header Anomaly to Network Discovery and Security Scanning. Insecure, 2009.
Detection for Identifying Hostile Network Traffic, Florida Institute of
[22] M. Mahoney and P. Chan, An Analysis of the 1999 DARPA/Lincoln
Technology, Tech. Rep., 2001.
Laboratory Evaluation Data for Network Anomaly Detection, in Recent
[5] A. Qayyum, M. Islam, and M. Jamil, Taxonomy of statistical based Advances in Intrusion Detection, 2003, vol. 2820, pp. 220237.
anomaly detection techniques for intrusion detection, in Proceedings
[23] D. Kennedy, J. OGorman, D. Kearns, and M. Aharoni, Metasploit: The
of the IEEE Symposium on Emerging Technologies, Sept 2005, pp. 270
Penetration Testers Guide, 2011.
276.
[24] A. Orebaugh, G. Ramirez, J. Burke, and L. Pesce, Wireshark & Ethereal
[6] E. Eskin, A. Arnold, M. Prerau, L. Portnoy, and S. Stolfo, A Geometric Network Protocol Analyzer Toolkit (Jay Beales Open Source Security).
Framework for Unsupervised Anomaly Detection, in Applications of Syngress Publishing, 2006.
Data Mining in Computer Security, 2002, vol. 6, pp. 77101.
[7] A. K. Ghosh, A. Schwartzbard, and M. Schatz, Learning Program
Behavior Profiles for Intrusion Detection, in Proceedings of the 1st
Conference on Workshop on Intrusion Detection and Network Monitor-
ing - Volume 1, 1999, pp. 66.

Potrebbero piacerti anche