Design of Soft Computing based Black Hole Detection in MANET
D. Vydeki
1
, K. S. Sujatha
1
and R. S. Bhuvaneswaran
2
1
Easwari Engineering College, Ramapuram, Chennai, India
2
Anna Univesity, Chennai, India
Keywords: Intrusion Detection, Black Hole, Fuzzy Logic, Genetic Algorithm.
Abstract: Mobile Ad hoc Networks (MANETs) are more vulnerable to attacks than the generic wireless network due
to the lack of an underlying infrastructure and shared medium. Intrusion Detection System (IDS) provides
an enhanced level of security to such networks. Application of Soft computing techniques to the detection
process is proved to be more suitable as they have human-like decision-making capabilities. This paper
proposes a hybrid intrusion detection system for MANETs that detects black hole attack, by combining
anomaly and specification-approaches of IDS. The proposed system aims at designing two different IDS
using the two fundamental soft computing mechanisms such as, Fuzzy and Genetic Algorithm (GA). The
design of each IDS is tested with simulated MANETs for various traffic conditions. The performance of
each system is compared based on the true and false positive rates. The experimental results show that the
fuzzy based system produces 81.8% true positive rate and the GA based system results in a 100% efficient
detection.
1 INTRODUCTION
A MANET is a collection of mobile platforms or
nodes where each node is free to move about
arbitrarily (Perkins, 2001). It is a mobile, wireless,
multi-hop network that operates without the benefit
of any existing infrastructure except for the nodes
themselves. Such networks are assumed to be self-
forming and self-healing. Routing in such networks
is challenging because typical routing protocols do
not operate efficiently in the presence of frequent
movements, intermittent connectivity, network splits
and joins. Moreover, use of wireless links make
these networks very vulnerable to security attacks
ranging from passive eavesdropping to active
interfering.
Various cryptography algorithms and secure
routing protocols attempt to prevent attacks to
provide the first layer of defense. An additional layer
of defense called “Intrusion detection” is often used
to protect networks. There are three major
approaches to IDS: misuse detection, anomaly
detection and specification-based detection. This
paper aims at developing a hybrid IDS that
combines the latter two methods using
computational intelligence techniques such as Fuzzy
logic, Genetic algorithm and Neural networks. By
developing a combinatorial approach, we get the
benefits of both methods that reflect in the detection
process.
This paper is organized as follows: Section 2
briefs about the related research work. : Section 3
provides an overview of IDS, AODV and the black
hole attack. The design of proposed systems and
their performance analysis is dealt in detail in
Section 4. Section 5 briefs about the future work and
conclusion.
2 RELATED WORK
In the recent past, plethora of research works has
been carried out in IDS. Ming-Yang Su has designed
IDS in which the three types of wireless nodes are
defined: normal, malicious and IDS nodes (Ming,
2011). A manually defined threshold is used in their
ABM (Anti-Blackhole Mechanism) function that
determines whether a node is a normal, suspicious or
black hole node. When a suspicious value exceeds a
threshold, a nearby IDS node will broadcast a block
message, informing all nodes on the network, to
cooperatively isolate the malicious node.
Anup Goyal and Chetan Kumar have suggested a
machine learning approach with GA, to identify
283
Vydeki D., S. Sujatha K. and S. Bhuvaneswaran R..
Design of Soft Computing based Black Hole Detection in MANET.
DOI: 10.5220/0004070802830288
In Proceedings of the International Conference on Signal Processing and Multimedia Applications and Wireless Information Networks and Systems
(WINSYS-2012), pages 283-288
ISBN: 978-989-8565-25-9
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
harmful/attack type of connections (Anup, 2010).
The algorithm takes into consideration different
features in network connections such as type of
protocol, network service on the destination and
status of the connection to generate a classification
rule set. For this experiment, they have implemented
a GA and trained it on the KDD Cup 99 data set to
generate a rule set that can be applied to the IDS to
identify and classify different types of attack
connections.
3 IDS, ROUTING AND ATTACK
Intrusion detection technologies focus on detecting
malicious activity typically from attackers that have
successfully penetrated the perimeter defences.
Based on the techniques used cross the security
barrier, IDS can be classified into three main
categories as follows:
i. Misuse Detection: In misuse detection, decisions
are made on the basis of earlier knowledge of the
intrusive process and what traces it might leave on
the observed system (Anjum, 2007). Such a system
tries to detect intrusion irrespective of any
knowledge regarding the background traffic. There
are several approaches in the signature detection,
which differ in representation and matching
algorithms employed to detect the intrusion patterns.
ii. Anomaly Detection: This technique establishes a
“normal activity profile” for a system and flags
observed activities that abnormally deviate from the
recognized normal usage as anomalies. It must first
be trained using normal data before it can be
released in an operative detection mode. The main
advantage of this model is that it can detect
unknown attacks. On the other hand, its
disadvantage is that it has high false positive alarm
rate when normal user profiles, operating system, or
network behavior vary widely from their normal
behavior.
iii. Specification-based detection: Specification-
based detection defines a set of constraints that
describe the correct operation of a protocol, and
monitors the execution of protocol with respect to
the defined constraints. This technique may provide
the capability to detect previously unknown attacks,
while exhibiting a low false positive rate.
The Ad hoc on demand distance vector routing
protocol (AODV) is a popular unicast routing
protocol that provides quick and efficient route
establishment between nodes desiring
communication with minimal control overhead and
minimal route acquisition delay (Sivaram, 2007).
Route discovery with AODV is purely on demand
and follows a route request/ route reply discovery
cycle. Requests are sent using a Route Request
(RREQ) message. Information enabling the creation
of a route is sent back in a Route Reply message
(RREP). In case where the source node receives
multiple RREP messages, it will select a RREP
message with the largest destination sequence
number value.
AODV is efficient and scalable in terms of
network performance, but it allows attackers to
easily advertise falsified route information to
redirect routes and to launch various kinds of
attacks. In each AODV routing packet, some critical
fields such as hop count, sequence numbers of
source and destination, and RREQ ID, are essential
to the correct protocol execution. Any misuse of
these fields can cause AODV to malfunction.
One such is misuse is the black hole attack which
targets on tampering the hop count and sequence
number fields. A node, which poses this attack,
changes the hop count value to 1 and the destination
sequence number to the largest value. This makes
the attacking node to be selected in the route
discovery process. The black hole node then
participates in the communication from source to
destination. When it receives packets from source, it
does not forward them to the intended destination;
instead, it drops them, thus disrupting the network
operation.
In this paper, the anomaly and specification
based techniques are combined, and tested on
networks that operate on AODV to detect black hole
attacking nodes, by making use of the advantages of
the two techniques to improve the detection rate.
4 SOFT COMPUTING BASED IDS
Soft computing is a multidisciplinary system defined
as the fusion of fields of Fuzzy logic, neuro
computing, genetic computing and probabilistic
computing. Soft computing is designed to model and
enable solutions for real world problems, which are
difficult to do using mathematical techniques.
Fuzzy inference is the process of formulating the
mapping from a given input to an output using fuzzy
logic. The mapping then provides a basis from
which decisions can be made, or patterns discerned.
Fuzzy inference systems have been successfully
applied in data classification, decision analysis, and
expert systems. There are two types of Fuzzy
Inference Systems (FIS):
WINSYS 2012 - International Conference on Wireless Information Networks and Systems
284
1. Mamdani Type, and
2. Sugeno Type.
This paper makes use of Sugeno type FIS because it
works well with optimization and adaptive
techniques.
GA is a family of computational models, which
uses the principles of evolution and natural
selection. These algorithms convert the problem in a
specific domain into a model by using a
chromosome like data structure and evolve the
chromosomes using selection, recombination and
mutation operations. In computer security, it is
mainly used to find an optimal solution to a specific
problem.
Many intrusion detection systems using either
signature or anomaly techniques are discussed in
literature. A Specification based system is also
designed for AODV using finite state machine
(Seng, 2003). Some hybrid approaches combining
misuse and anomaly are found in the literature.
Here, it is proposed to combine the anomaly and
specification based methods to improve the
detection rate and to reduce the false positives. The
anomaly technique detects the malicious nodes by
describing the normal behaviour. Specification
based mechanism crafts the rules that are related to
the protocol. These two features are combined in the
proposed design as follows. The simulated networks
consisting of normal nodes and black hole nodes are
run with various traffic conditions. The volume of
traffic varies broadly to provide data in the low,
medium and heavy traffic categories. This is the
training mechanism of anomaly system. From the
available data set, the features related to AODV
protocol are extracted, which is a typical
specification based approach. In addition to
combining these two methods, the system is tested
the using the two basic computational intelligence
methods. The results for the two systems using FIS
and GA are compared.
The algorithm of the proposed system is outlined
as follows:
1. From the training data set, the features that are
relevant to AODV are extracted. They are:
Number of packets dropped (PD) which
dictates the behaviour of the black hole. Simply
because a node drops more number of packets
compared to any other node in the network, it
does not lead to the decision of black hole node.
This may be due to any of the reasons mentioned
in assumption numbered 2.
Number of RREPs sent by any node (SREP)
is greater for a black hole node for the simple
reason that the adversary node tries to make itself
available in the communication path.
Number of RREQs (RFR) forwarded will
indicate the malicious behaviour of the black
hole node as it does not forward any RREQ to
the neighbouring node to avoid other nodes
participating in the RDP.
2. As dictated by the specification-based approach,
the threshold value for the selected parameters are
defined, but using the anomaly technique, as the
mean of each feature.
3. To enhance the detection process a credit
allocation scheme is introduced in the system. Each
node is allotted an initial credit, which is determined
by the RFR. The credit is incremented or made to
zero as per the following rule:
If the value of each parameter at each node is
lesser than the respective threshold value, the credit
is incremented by one.
Else the credit is assigned zero.
4. The total credit (TC) of each node is the sum of
individual credits.
5. The parameter values of all nodes, the respective
credit values are given as inputs to the FIS and GA
systems.
6. The detailed algorithm used in each of the
system is described in the later sections.
A novel combinatorial design approach of IDS that
uses the anomaly and specification based technique
is proposed. Also the design is tested using the two
basic soft computing mechanisms. Generally, an
IDS consists of the following fundamental
components: Data collection components, Data-
analysis components.
In this system, data collection is carried out using
ns2, which is a network simulation tool. MANETs
with 15, 25 and 50 nodes are created with AODV as
routing protocol. Special data collecting (DC) nodes
are placed in a strategic position so as to do the data
collection process. These nodes move throughout the
network and work in promiscuous mode. Hence they
are capable of monitoring the behaviour of each
wireless node in its vicinity. To simulate black hole
attack, the hop count and the sequence number in the
RREP is appropriately modified in the AODV, the
changed protocol is named as ‘Black hole AODV’
(BAODV). The nodes that use this BAODV are
treated as black hole nodes. Many networks with a
mixture of genuine and black hole nodes are
simulated. The number of black hole nodes varies
from 1 to 5% of the total nodes in the network. One-
to-one communications between a variable numbers
Design of Soft Computing based Black Hole Detection in MANET
285
of pair of nodes are simulated with random walk
mobility model. Following are the assumptions
made:
1. All the nodes in the network are genuine in
forwarding the RREQ packets to the neighbouring
nodes in the coverage region, when they don’t have
a route to the destination defined in the request
packet.
2. Nodes may drop the received data packet only
under the following two conditions: a). The network
is severely congested due to heavy traffic, or, b).
The source/destination node or the intermediate
node on the identified route had moved out of range
of the forwarding node.
3. A genuine node may even try to establish
communication with the black hole node.
4.1 FIS based IDS
The IDS that uses FIS works with Sugeno type as it
does not require defuzzification process. Figure 2
shows the basic operations of the FIS. The input and
credit values from the database are grouped into
various clusters. Either of the following two
methods is used for the purpose:
1. Subtractive Clustering
2. Fuzzy C-means Clustering (FCM)
Subtractive clustering is a fast one pass algorithm
for estimating the number of clusters and the cluster
centers in a set of data. In FCM, each data point
belongs to a cluster to some degree that is specified
by a membership grade. Both clustering mechanisms
were incorporated in the proposed system and the
subtractive clustering mechanism is found to be
more appropriate for the given application.
Figure 1: Sugeno FIS.
The input and output clusters are defined interms
of membership functions (MFs). A MF is a means
through which each point in input value is mapped
to a membership value between 0 and 1. Gaussian
distribution has been selected as the MFs for this
system. This process is depicted as ‘defining MF’ in
figure 1. After clustering, and mapping the input and
output values to the appropriate MFs, various rule
sets are applied on them. Sugeno type FIS defines its
own set of rules based on the clustering. Finally
defuzzified output produced by the FIS consists of
values that give an indication to the performance of
each node. The entire process is shown in figure 1.
The algorithm for the FIS based IDS is as given
below:
Get input: PD, SREP, RFR, TC
Perform the clustering operation using
subtractive and FCM methods.
Define the Gaussian MF for each cluster of
inputs.
Generate a Sugeno-type 2 FIS to apply the rule
set on the inputs.
From the output of FIS, compute the standard
deviation to facilitate the detection process.
Compare the performance of each node with
respect to the calculated standard deviation as
described below:
If the FIS output of the node is lesser than the
standard deviation, it is considered to be a
normal node.
If the FIS output of a node is greater than the
standard deviation, it is detected as black hole
node.
The nodes that are identified as black hole nodes
are displayed with their respective node ids.
4.2 GA based IDS
The algorithm for GA based IDS is as follows:
Assign the initial population as the network
parameters derived from the data set: PD, SREP,
RFR
Calculate the initial threshold based on the
network parameters as follows:
T1 = Average ( NPi),
Where, T1 = First Threshold
NPi = Network Parameter , i = 1,2,3….N.
N= Total no of nodes in the network
Encode the chromosome based on the threshold
criterion and determine the chromosomes that are
above the threshold.
Shortlist the chromosomes based on their fitness.
Calculate the optimum parameter as per the
following condition:
If(NPi<=T1) then { NPi-op =1}
else {NPi-op =0}, Where NPi-op is the optimal
network parameter. This fitness evaluation process
WINSYS 2012 - International Conference on Wireless Information Networks and Systems
286
results in making the respective network parameter
value as ‘0’ for fit nodes (i.e., the black hole node)
and ‘1’ for unfit nodes (i.e., the normal nodes).
Select the survivor based on selection and
recombination criteria. If all the network parameter
values for any node is zero, it is selected as the
survivor black hole node.
Display the survivor node list along with its
respective node id.
4.3 Simulation Results & Performance
Analysis
This paper provides a combinatorial approach for
soft computing based intrusion detection. The design
was explained elaborately in the previous sections.
This section covers the simulation of the various
scenarios and the performance of each system. As
mentioned earlier, ns-2 was used to simulate various
MANETs that consist of different nodes. The black
hole nodes and DC nodes were simulated
appropriately. After the network had run with
various traffic and communication scenarios, the
parameters necessary for the proposed system were
derived from the data log of DC nodes. The various
details of the simulation are given in table 1.
Table 1: Simulation parameters.
Description
Value
Network Size
15, 25, 50
Max. Speed &
Pause Time
200m/s &
2s
No. of Black hole Nodes
1,2,3 (For 15 node N/w)
1,2,5 (For 25 node N/)
3,6,10 (For 50 node N/w)
Simulation Duration
500 s
Simulation Area
1000 1000 sq.m
Traffic Density
Low, medium, High
Type of Communication &
Application
One-to-One
CBR
The various parameters such as PD, SREP, RFR
and TC are given as inputs to the two systems. The
performance of each system is discussed later in this
section. The performance measure of any IDS can be
defined in terms of the following:
True Positive Rate (TPR)
False Positive Rate (FPR) and
True Negative Rate (TNR)
TPR is the measure of number of black hole nodes
correctly identified. FPR is the measure of number
of normal nodes identified as adversary nodes. TNR
indicates the number of black hole nodes being
detected as normal nodes. Our system based on the
three computational intelligence techniques are
analysed using the above three measures. The
performance of a FIS based IDS is given in figure 2.
0
2
4
6
8
10
14
-0.5
0
0.5
1
1.5
2
Node performance with 3 black holes,high traffic
Node Id
FIS Output
Figure 2: Detection using FIS based IDS in a MANET of
15 nodes with 3 black holes and high traffic condition.
From figure 2, it is inferred that the nodes with
the least value of performance are identified as black
hole nodes. Hence, the nodes having id 1, 13 and 14
are detected as black hole nodes by the FIS based
IDS. As per the simulation, the nodes with id 12, 13
and 14 are simulated as black holes. While our
system detects 13 and 14 correctly as black hole
nodes, it has not detected node 12, which is also a
black hole. It is also understood that node 1, which
is a normal node is being detected as an adversary
node. From the above result, it is clear that the FIS
based IDS produces true positives, true negatives
and also false positives. The results of detection
process of various MANETs using FIS are tabulated
in table 2.
From table 2, it is understood that in most of the
cases, the fuzzy based system produces 100% true
positive rate, which indicates its efficient detection.
Exactly the efficiency is 81.8%. On the flip side, it
results in false positives and false negatives. The
performance of our fuzzy system for 50 nodes is
plotted in figure 3.
Figure 3: Analysis for 50 nodes network.
The detection of malicious nodes using GA is
shown in figure 4. It can be easily deduced from
figure 4 that the survivor nodes are the black hole
nodes and they are displayed with their node ids. It
Design of Soft Computing based Black Hole Detection in MANET
287
is understood that the nodes with id 24 and 25 were
detected as adversary nodes. Earlier in the
simulation for this particular case consisted of nodes
24 and 25 as black hole nodes. Moreover, GA based
system does not produce any false positives or false
negatives.
Table 2: Analysis of FIS based IDS.
No.of
Nodes
No.of Black Hole
Nodes
Traffic
TPR
FNR
FPR
15
1
Low
100
0
11.3
Med
100
0
7
High
100
0
7
15
2
Low
100
0
11
Med
100
0
13.3
High
100
0
7
15
3
Low
100
0
11
Med
89
11
11.3
High
89
11
7
25
1
Low
100
0
20
Med
100
0
13
High
100
0
13
25
2
Low
100
0
11
Med
100
0
13
High
100
0
13
25
5
Low
100
0
13
Med
100
0
13
High
100
0
13
50
3
Low
55.3
44.7
24.6
Med
89
11
15.3
High
100
0
22.3
50
6
Low
100
0
24.6
Med
100
0
27
High
100
0
27
50
10
Low
100
0
24.6
Med
100
0
27
High
66.7
33.3
18
2 Black holes
0
0.2
0.4
0.6
0.8
1
1.2
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Node ID
Survivor
Survivor
Figure 4: Detection of GA based IDS for a network of 25
nodes and 2 black holes.
5 CONCLUSION & FUTURE
WORK
A soft computing based IDS that combine the
specification and anomaly approaches was
presented. The algorithms using computational
intelligence techniques were developed to detect
black hole attacks in MANETs The two fundamental
soft computing techniques such as fuzzy and genetic
algorithm were used. Simulation study of these
systems was also exhibited. Among the techniques
GA based technique has 100% detection rate
compared to the 81.8% detection rate in FIS based
system. In future, publishing the identity of the
detected node throughout the network using the DC
nodes will be carried out. Our future work includes
improving the detection rate of fuzzy based system
using cross layer approach. Also the possibilities of
this system against other type of attacks will be
explored.
REFERENCES
Charles E. Perkins, 2001, Ad hoc Networking”, Addison-
Wesley.
Farooq Anjum, 2007, Security for Wireless Ad hoc
Networks”, John Wiley & Sons.
Ming-Yang Su, 2011, “Prevention of Selective black hole
attacks on mobile attack networks through IDS”,
Computer Communications.
Anup Goyal and Chetan Kumar, 2010, “ GA-NIDS: A
Genetic Algorithm based Intrusion Detection System”.
C.Sivaram Murthy, B. S. Manoj, 2007, Ad hoc Wireless
Networks, Protocols and Architecture”, Pearson
Education.
Chin-Yang Tseng, 2003, Poornima Balasubramanyam,
Calvin Ko, “A Specification-based Intrusion Detection
System for AODV”, Proceedings of the 1st ACM
workshop on Security of ad hoc and sensor networks.
WINSYS 2012 - International Conference on Wireless Information Networks and Systems
288