AN INDEPENDENT TRUST MODEL FOR MANET BASED ON FUZZY LOGIC RULES

http://www.iaeme.com/IJARET/index.asp 278 editor@iaeme.com
International Journal of Advanced Research in Engineering and Technology (IJARET)
Volume 10, Issue 2, March-April 2019, pp. 278-289. Article ID: IJARET_10_02_028
Available online at http://www.iaeme.com/IJARET/issues.asp?JType=IJARET&VType=10&IType=02
ISSN Print: 0976-6480 and ISSN Online: 0976-6499
© IAEME Publication
AN INDEPENDENT TRUST MODEL FOR
MANET BASED ON FUZZY LOGIC RULES
Syeda Kausar Fatima
Research Scholar JNTUH, Prof., DCET, Hyderabad, Principal NSAKCET,
Prof., ECE Dept, JNTUH, India.
Dr. Syeda Gauhar Fatima
Dr. Syed Abdul Sattar
Dr. D. Srinivasa Rao
ABSTRACT
A mobile ad hoc network (MANET) is a self-organized system comprised by multiple
mobile wireless nodes. Due to the openness in network topology and the absence of
centralized administration in management, MANET is vulnerable to attacks from
malicious nodes. In order to reduce the hazards from these malicious nodes, we
incorporate the concept of trust into the MANET, and build a subjective trust
management model with multiple decision factors based on the analytic hierarchy
process (AHP) theory and the fuzzy logic rules prediction method ˉ AFStrust. We
consider multiple decision factors, including direct trust, recommendation trust,
incentive function and active degree, in our model to reflect trust relationship’s
complexity and uncertainty from various aspects. It overcomes the shortage of
traditional method, where the decision factors are incomplete. Moreover, the weight of
classification is set up by AHP for these decision factors, which makes the model has a
better rationality and a higher practicability. Compared to the existing trust
management models, comprehensive experiments have been conducted to evaluate the
efficiency of our trust management model in the improvement of network interaction
quality, trust dynamic adaptability, malicious node identification, attack resistance and
enhancements of system’s security.
Keywords: MANET; Trust Management Model; Decision Factors; Analytic Hierarchy
Process; Fuzzy Logic Rules

Syeda Kausar Fatima, Dr. Syeda Gauhar Fatima Dr. Syed Abdul Sattar and Dr. D. Srinivasa
Rao
Cite this Article Syeda Kausar Fatima, Dr. Syeda Gauhar Fatima Dr. Syed Abdul Sattar
and Dr. D. Srinivasa Rao, An Independent Trust Model for Manet Based on Fuzzy Logic
Rules, International Journal of Advanced Research in Engineering and Technology,
10(2), 2019, pp. 278-289.
http://www.iaeme.com/IJARET/issues.asp?JType=IJARET&VType=10&IType=2
1. INTRODUCTION
A mobile ad hoc network (MANET) is a self-organized multi-hop system comprised by
multiple mobile wireless nodes with peer-to-peer relationships. The nodes in the network
cannot communicate with each other via well-established infrastructure. Due to the limitation
of energy, two peers out of communication range require intermediate nodes to transfer
messages. Therefore, a node in this network serves as a host and a router simultaneously. Each
node is assumed to relay packets for other nodes, and it works well only if the nodes in the
network behave cooperatively. Due to the openness in network topology, MANET often suffers
from attacks by selfish or malicious nodes, such as the on-off attack, bad-mouthing attack,
conflict behavior attack, packet dropping (black-hole) attack, selective forwarding (gray-hole)
attack and so on [1]. Existing security technologies are mostly based on encryption and
authentication, which are unsuitable in the dynamic network topology without a trusted third-
party. Moreover, the traditional cryptosystem-based security mechanism is typically used to
resist the external attacks. They show inefficiency in handling the attacks from the internal
malicious nodes which may lead to serious influence on the security, the confidentiality, and
the life cycle of the whole network.
Trust management mechanism is considered to be an effective measurement to solve these
problems [2]. In the context of MANET, there are several trust management models that have
been proposed in the realm of network (e.g., [3, 4, 5, 6]), where trust can be considered as the
reliance of a network node on the ability to forward packets or offer services timely, integrally
and reliably. In the existing models, decision factors are often incomplete in the trust derivation,
which are not fully integrated with the inherent characteristics of MANET. When the factors of
decision-making are given, though we know that different factors have different weights, the
precise weights are difficult to determine. Existing methods in these models for weight
determination are lack of rationality and practicability. As a result, they cannot calculate an
accurate trust value for each node. Hence, these models are ineffective in MANET trust
management, and their applications are very simple.
To address those questions, in this paper, we establish a new subjective trust management
model for MANET considering the behaviors of the dynamic nodes in the open environment
and the complete decision factors of nodes’ trustworthiness. The nodes’ trust values can be
easily used in trust management strategy, which includes the applications anti-attack, decision
making etc. The motivations of our work are to (a) obtain a more accurate node’s trust value;
(b) improve the quality of network interaction, increase the proportion number of good
recommendation, raise the malicious node’s correct detection ratio; (c) decrease the hazards
from these malicious nodes and protect the network from internal attacks (e.g. mitigate
cooperative denigration attacks); (d) enhance the network’s ability of trust decisions (e.g.,
trusted routing decisions).
The remaining paper is organized as follows. Section 2 discusses the related work. In Section
3, we describe our trust management model, and the calculation of trust value is presented in
Section 4. Section 5 presents the experimental results on the performance of our trust model.
Finally, Section 6 gives the concluding remarks along with extensions and directions for
future research.

An Independent Trust Model for Manet Based on Fuzzy Logic Rules
2. RELATED WORK
In the context of MANET, there are multiple trust management models that have been proposed
in the realm of network. In MANET, trust can be considered as the reliance of a network node
on the ability of other nodes to forward packets or offer services timely, integrally and reliably.
From the evolutionism and sociology points of view, Mui [3] et al. firstly introduced a trust
and reputation computation model for generalized networks. In the indirect trust evaluation
process, they proposed a graph parallelization algorithm, which is intuitive and easy to
understand. Based on the work of Mui, Durad et al. introduced a new term: trust of scaling
factor (TSF2) [4], emphasizing the contribution of direct interactions and the rationality of
recommendation. They also proposed a modified transformation algorithm (MTA) for TSF2
calculation.
Using the theory of semi-rings, George et al. proposed a new trust model for ad hoc
networks [5]. In the model, they described trust evaluation scheme as a routing path problem in
a directed weighted graph. When gathering the opinion that one entity has about another entity,
they defined two binary operations to evaluate opinion values from single and multiple
recommendation paths respectively.
Luo et al. [6] proposed a subjective trust management model based on certainty-factor for
MANET (CFStrust) after considering fuzzy set theory and reputation model, which can be used
to quantify and evaluate the nodes’ credibility. In this model, the problem of trust management
is modeled by fuzzy likelihood estimation and confidence estimation. The trust evaluation
mechanism and the derivation rules of recommendation trust relationship are given in this
model. Although two effective factors corresponding with mathematical derivation are
discussed, it does not take a comprehensive account of the nodes’ computing power, the
instability of information transmission through multi hops and trust’s attenuation problem etc.
3. TRUST MANAGEMENT MODEL
In ad hoc networks, every node act as host and router simultaneously. Trust is a relationship
between two neighbor entities. Trust value expresses the degree that one node expects another
node to offer certain services.
A. Overview Our Trust Model
An ad hoc network is always comprised of many entities, and each entity is an independent
node. In this section, we present our trust model from graph theory, which is denoted as M=<V,
E, f>. (1) Trust entity set can be defined as V= {v1, v2, …, vn}, where n is the scale of the
network; (2) E is a relation on V, and |E| is the number of directed network links. Each eij in E
represents a directed edge from node vi to its neighbor node vj; (3) f: f e (ij ) → ∈R [0,1]denotes
the trust value
(a real number between 0 and 1) of each edge eij.
According to our definition, the trust model of an ad hoc network can be represented as a
directed weighted graph. For example, in Figure 1, there are five nodes in this ad hoc network.
Each dashed circle represents the radiation scope of the corresponding node, where the nodes
within the scope are neighbor nodes that can communicate directly.

Rao
Figure 1 Trust network graph in Ad hoc networks
B. Nodes’ Trust Levels
In our model, TV denotes for a node’s trust value, which is defined in a continuous range
between 0 and 1 (i.e. 0” TVij” 1). Let vi and vj represent the evaluating and evaluated nodes,
respectively. The trust value 0 signifies complete distrust, while the value 1 implies absolute
trust. We define simple grading criteria for trust, and an example of node’s trust levels is listed
in Table 1. A threshold value Ș, termed as the black-list trust threshold, is used to detect
malicious nodes. In other words, if the trust value of a node is smaller than Ș, it will be regarded
as a malicious node by its evaluating node.
Table 1 Trust levels of nodes
Level
Trust
Value
Meaning
1 [0, Ș) Malicious node
2 [Ș, 0.7) Low trustworthy node
3
[0.7,
0.9)
Trustworthy node
4 [0.9, 1] Complete trustworthy node
C. Node’s Trust Table
Each node in our model additionally owns a trust table (Table 2 which bonds with Figure 1)
with items defined as follows.
Table 2 Node v1’s trust table
Nb Tin Tout Black-List
v2 0.90 0.92 No
v3 0.79 0.23 Yes
In each row of the table, Nb denotes node v1’s neighbor that can communicate with v1 via
a single-hop; Tin is the trust value that the neighbor node gets about node v1; Tout is the trust
value that node v1 has about the neighbor; Black-List indicates whether v1 considers this
neighbor as a malicious node (e.g., the black-list trust threshold Ș as discussed in Table 1 is set
to 0.4 in this example).
D. Classification of Trust Types
In AFStrust model, there are three types of trust value, which are historical trust value, current
trust value and path trust value.

1) Node’s historical trust: it is estimated by the node’s physical neighbors based on
historical interaction information at the end of each time interval, which is calculated by four
decision factors. These factors are direct trust denoted by DTij, recommendation trust denoted
by RTij, incentive function denoted by IFij and active degree denoted by ADij. The historical
trust value HTVij denotes node vj’s trust level from the evaluating node vi’s point of view,
which can be computed by:
HTVij =α β γ μDTij + RTij + IFij + ADij (1)
Where Į, ȕ, Ȗ, ȝ are coefficients, such that
(αβγ μ+ + + =1). The values of Į, ȕ, Ȗ, ȝ are decided by the influencing on the assessment
of trust value in different environment.
2) Node’s current trust as shown in Figure 2, a node’s current trust value predicts the
node’s trust value in the next time interval t+1. In our proposed model, it is computed from the
node’s historical trust value based on the fuzzy logic rules prediction method. In this paper, we
use the term ‘trust value’ for a node’s current trust value, for simplicity of representation.
Figure 2 The types of trust value transfer graph
3) Path trust: it expresses the credibility for the set of nodes on a routing path, and its value
is defined as the minimum of single-hop trust values. The service (source) nodes determine
service level basing on the assessment of path trust value. As a result, the path trust value can
be defined as a constraint in the trusted routing decision making.
Path trust: PathTVij = min {TVmk (2)
≤ ≤ −
From Figure 1, we easily see the examples of nodes’ trust values and path trust values:
TV12=0.92, TV45=0.83, TV43=0.19; path trust value: PathTV14=Path (1 ˉ >2 ˉ>4) =min
{0.92, 0.95} =0.92. Setting Ș=0.4, we see that, node 3 exists in the black lists of all its neighbors,
which is regarded as a malicious node by node 1, 2, 4 and will be excluded from the local
network for a special time.
4. CALCULATION OF NODE’S TRUST VALUE
Trust evaluation is the core of trust management system, including the trust definition, trust
synthesis, and trust update.
A. Calculation of Historical Trust Value
Those mentioned trust management models analyze trust’s decision factors and give different
models to calculate the weights of these factors.
1) The trust decision factors
There are multiple decision factors to assess a node’s trust value in existing trust models [7,
8]. For example, two neighbor nodes in an ad hoc network, which can interact with each other
directly, may establish a direct trust relationship. Two nodes can also establish relationship by
intermediate node’s (or path’s) recommendation, which is usually called indirect trust or

Rao
recommendation trust. Besides the above two factors, we introduce another two decision
factors, which are incentive function and active degree into our trust model.
a) Direct trust evaluation rules
In our model, we consider the interaction which occurs between a node and its physical
neighbor as a direct interaction, of which the evaluation is defined as direct trust evaluation;
while the interaction via multi-hops is defined as indirect interaction, of which the evaluation
is defined as indirect trust evaluation (Involved in the following subsection). According to the
two special interaction sides (e.g., nodes vi and vj), node vi make a satisfaction evaluation of
each direct interaction interacted with its neighbor node vj, denoted by f(i,j) (0” f(i,j)” 1). After
evaluation step, the results are obtained in evaluating node’s (i.e., vi’s) local memory. In order
to obtain an accurate node’s trust value, our model distinguishes the different influence of each
interaction interval. Using the time stamp mechanism to analyze each interaction interval (e.g.
set interval ǻt =30(s)). Till the current interaction time, there will be n intervals [t1, t2… tn] in
simulation system. For the k-th interaction interval, there are Ntk number of interactions, node
vi makes a final direct trust evaluation for node vj
Definition 1 Time decay function: the attenuation rate function which is made by the k-th
interaction interval compares to the latest interaction interval in the trust’s calculation is defined
as the time decay function.
fk =ρn k− ,0 < <ρ 1,1≤ ≤k n (3)
The base coefficient ȡ represents the attenuation factor. A smaller ȡ causes a greater
attenuation of fk, and vice versa.
Finally, node vi calculates a direct trust value for node vj according to their history
interaction evaluations.
Regulatory factor σ is used to scale the impact of number of interactions on the direct trust
computation, and the interaction factor is denoted by. The concrete value of σ can be adjusted
based on the environment and characteristics of the application. The interaction factor has a
negative exponential growth to the number of interactions in a given interval. This factor is
used to emphasize the importance of the transaction number. The direct trust value is the
weighted average of all interaction evaluations in different interaction intervals.
b) Recommendation trust evaluation rules
Definition 2 Recommending credibility: it represents the credibility degree of the
recommending node (or recommending path) to provide recommendation experience, which is
denoted by RC.
We set a recommending credibility threshold d1, if a recommending node’s recommending
credibility is less than this threshold (RC<d1), the recommendation experience provided by this
node is not considered.
Suppose node vi gets indirect recommendation trust value of node vj by the intermediate
recommendation of node vm (a single node). Setting DTim = p, DTmj = q, we can calculate:
RTimj* = RCim × DTmj = DTim ×DTmj = p q* (4)
This can be shown in Figure 3(a).

Figure 3 Node’s Expurgating Rules
Suppose node vi gets a recommendation trust value of node vj by the intermediate
recommendation of path P.
Setting DRiPathP = p, Rnj = q, then we have:
RTiPathj = RCiPathP ×Rnj = p q* (5)
Node vn is the Path P’s terminal node. This can be shown in Figure 3(b). The single
intermediate node’s recommendation style can be seen as a special form of path’s
recommendation.
Assume node vi gets recommendation experience RT*ij by trust propagation along a
directed recommending path P=<vi, v1, v2, …, vn, vj>. Currently we can’t say the value RT*ij
got from trust attenuation rules is a reasonable value, because the recommending nodes in the
path is not absolutely trustworthy from node i’s point of view. In recommendation experience
evaluation rules, the node vm’s recommending credibility in this recommending path is denoted
by RCim.
RCim = DTi1 *DT12 *...*DT(m−1) m (6)
Capturing the notion of social networks, node vi should give an objective estimate to all
recommending nodes in a recommending path. We obtain a recommending credibility value
RCiPathP for path P’s recommending credibility,
Where RCikv is the recommending credibility that node vi has about node vk. Therefore,
the path P makes recommendation experience as RTiPj:
RTiPj = DRiPathP *Rnj (7)
Node vn is path P’s terminal node.
If there are n recommending paths P1, P2, …, Pn between node vi and vj, node vi should
give an objective estimate to all recommending paths. Each path has its local weight
recommending credibility WPath. However, when comes to all, each path plays different role
which based on their local weight. Then node vi should give an objective estimate for each path.
Assuming the in-degree of node vj is n, it means there are n paths from node vi to node vj.
Using trust attenuation rules, we can get n recommendation experience: RTiP1j, RTiP2j, …,
RTiPnj. In terms of recommending path weighting rules for multiple paths above, we can
calculate these recommending paths’ general weights and get WPk (kę {1, 2, n}).
c) Incentive function evaluation rules
Definition 3 Incentive function: this function reflects the incentive for cooperative entities.
Because of the cooperative entities often have fewer bad interactions and less interaction
failure rates, while malicious or uncooperative entities often refuse to or interrupt service.
Incentive function also reflects that the system would make some punishment to the
uncooperative entities. This function is denoted by IFij, which is calculated with following
equation:
IFij = −1 φ () n (8)

Rao
Ɏ(n) represents penalty factor, which is used to indicate that the node does not fulfill its
responsibility and do harm to evaluating a node’s trust value. When the satisfaction evaluations
of interactions are less than a threshold d, we call those interactions as malicious interactions.
Till the current interaction time (T) between node vi and node vj, the total number of malicious
interactions is denoted by mT, the total number of interactions is denoted by NT. The penalty
factor is calculated with the following equation:
NT−mT
φ ϕ ϕ () n = NT (<1) (9)
d) Active degree evaluation rules
Definition 4 Active degree: this decision factor reflects the level of activity of an entity in a
network.
It is used to indicate the credibility of evaluated entity. If an (evaluated) entity has a higher
active degree, other (evaluating) entities is willing to interact with it due to its expected higher
trust level. An evaluating node vi records the cumulative number of entities interacting with an
evaluated node vj, and calculates the active degree of the evaluated node as follows:
ADij = −1 η, L ≥ 0
L represents for the cumulative number of entities interacted with the evaluated node vj. Ș
termed as the black-list trust threshold (discussed in Subsection 3.2).
2) AHP theory
Analytic Hierarchy Process (AHP) [9] theory was proposed by Saaty in the 1970s, which has
become one of the essential multi-criteria decision-making methods. It combines qualitative
and quantitative factors in the analysis via a multi-index synthetic assessment algorithm. The
analysis can be divided into the following four steps: (1) construct a hierarchical structure
model; (2) form basic judgment matrix; (3) calculate the weight vector of decision factors; (4)
make a consistency test for the judgment matrix.
We use AHP to precisely determine the weights of trust’s decision factors based on the
nodes’ historical behaviors. Node’s historical trust value:
HTVij=0.56DTij+0.264RTij+ 0.122IFij+0.054ADij
B. Calculation of Current Trust Value
When the requested (transmitter) node receives a packet transmission request, it’s hard for
requesting node to evaluate whether the requested node is willing or not to provide the service.
However, its historical interactions can be recorded and the node’s capability level can be
monitored, therefore, we can model these factors as follows [10]: Let TV(t) represents a
evaluated (requested or transmitter) node’s historical trust level at the end of time interval t, its
record of historical behaviors on offering certain services in the latest time interval, which has
been measured in Subsection 4.1 using AHP theory. Let C(t+1) represents the same node’s
capability level on providing service level for the next time interval t+1 (prediction time
interval), which includes the remnant utilization ratio of battery, local memory, CPU cycle, and
bandwidth at that point. Let TV(t+1) refers to the same node’s current trust level for the next
time interval t+1. Assume the fuzzy membership function of TV(t) or TV(t+1) consists of four
fuzzy sets VeryLow (VL-Malicious node), Low (L-Low trustworthy node), Medial (M-
Trustworthy node) and
High (H-Complete trustworthy node), and the fuzzy member function of C(t+1) consists of
three fuzzy sets LOW (L-Low capability level), Medial (M-Medium capability level) and High

(H-High capability level), respectively. Combined with social control theory, we give the fuzzy
inference rules as follows (Table 3):
Table 3 Logical rules prediction on trust levels
The rules in the above table actually establish a mapping function from TV(t)×C(t+1) to
TV(t+1), which is based on the analysis of the node’s historical behaviors and current
conditions. For example, when an overloaded node lacks the CPU cycles, buffer space or
available network bandwidth to forward packets, with such a low capability level, even if its
historical trust level is very high, it’s also untrustworthy in next time. This only shows the first
rule from above table. Corresponding with each rule, there is an inference relationship Rl :
Rl = TVt ×Ct+1×TVt+1
That is for ∀ ∈h TV t (), c∈C t (+1), u ∈TV t (+1), we have
Rl (h c u,) = TV h () ∧C c () ∧TV u ()
For each pair of given (TV t () *, C t (+1) *), using the general total relationship R, we can
obtain the output:
TV t (+ =1) *(TV t () *×C t(+1) )* R
Then with the help of the maximum membership degree approach, we can get an explicitly
node’s current trust value (a complex representation for node’s trust value) u*∈ [0,1] by
defuzzification. For each time interval (ǻt), at the end of sub-time interval (e.g. one fifth of the
time interval, ǻt/5), making the latest node’s current trust value as node’s historical value and
taking this value as the input, we recycle the fuzzy logic rules prediction method to update the
node’s rust value.
5. EXPERIMENTAL RESULTS
All experiments are carried out on a PC machine with a Pentium 4 processor (2.4 GHz) and
2GB main memory. To evaluate the performance and validity of Durad’s TSF2 [4], Luo’s
CFStrust [6] and our management model AFStrust, we have conducted a comprehensive test
using Netlogo simulator [11].
A. Experiment Setup
We use Netlogo platform to construct a simulated environment for MANET and make a
comparative analysis for the performance of the related models. The platform initializes a
MANET with 40 nodes, which contains malicious nodes. Basic experimental parameters are
set as follows [12].
Table 4 Parameters setting
N T ǻt ȡ ı ij d η
40 300(s) 30(s) 0.9 5 0.9 0.7 0.4
In Table 4, the network size denoted by N, the simulation time denoted by T, the duration
of times tamp denoted by ǻt, the base number of decay function denoted by ȡ, the adjustable
parameters of the factor for the number of transactions denoted by ı, the base number of the
penalty factor denoted by ij, the threshold value for the evaluation of malicious interaction
denoted by d, and the black-list trust threshold denoted by Ș.
TE ( t )
C ( t +1)
HVL L M
LL VL
M VL M
H VL L M H

Rao
B. Contributions to Network
1) Ad hoc Network with 10% malicious nodes
The platform initializes a MANET which contains with 10% percentage of malicious nodes.
We make a comparison in terms of performance between TSF2, CFStrust and our trust model
as shown in Figure 4.
(a) (b) (c)
Figure 4 Simulating Ad hoc Network with malicious nodes
Figure 4(a) demonstrates the comparison of the satisfaction rates of network interaction
with no-model, TSF2, CFStrust and AFStrust models. From this figure, we see that the
satisfaction rates of network interaction with trust management models rise with the increase
of simulation time, while the satisfaction rate with no-model decreases gently, and the
satisfaction rates with trust models are significantly higher than that with no trust model. This
advancement with trust models can be attributed to the node’s trust mechanism, which elevates
the probability of successful (or good) service to a trustworthy node. If one node is thought to
be as a malicious node by its neighbors who will not continue the interactions with it, so the
interactions will be only happened among normal nodes, leading to that the satisfaction rates of
network interaction rise. While with no-model, malicious nodes longer exist, the greater
damage on the network, the satisfaction rate of network interaction reduces with the operation
of system. The performance with AFStrust is better than those with TSF2 and CFStrust. Taking
simulation time 120(s) for example, with no-model and three trust models (TSF2, CFStrust and
AFStrust), we get the satisfaction rates of network interaction as 0.81, 0.87, 0.88, 0.92
respectively. Comparing with CFStrust, the satisfaction rate based on AHPSLtrust is increased
by 4%, comparing with no-model that is increased by 11%.
The comparison of the numbers of good recommendation with different trust models are
shown in Figure 4(b). The curves of GoodnoModel, GoodTSF2, GoodCFStrust and
GoodAFStrust respectively represent the number of good recommendations in network with no
model, TSF2, CFStrust and AFStrust model. From this figure, we see that with the increase of
total number of recommendations in network, the proportion numbers of good recommendation
with trust models rise, while the number with no-model decreases gently. The proportion
number with trust model is significantly higher than with no-model. Due to the help of trust
management mechanism, the network could identify its inherent malicious nodes. When
calculating the nodes’ trust values, the proportion of good recommendation experience
provided by malicious nodes will be neglected, and the malicious nodes will be slowly removed
from the set of recommending nodes. The proportion number of bad recommendations gets
smaller with the operation of system, so good recommendation occupies total recommendation
ascend. Due to the detailed and complete recommendation trust evaluation mechanism in
AFStrust model, the performance with this model is better than TSF2 and CFStrust. Taking
simulation time 120(s) for example, the proportion number of good recommendations with
AFStrust is 5 percentage points higher than that with CFStrust.

From Figure 4(c), we see that the comparison of the correct detection ratio of malicious
nodes. Trust evaluation is an attractive target for those malicious adversaries. The results of
trust evaluation can be used to detect malicious nodes in a network. The curves of TSF2,
CFStrust and AFStrust respectively represent the correct detection ratio of malicious nodes with
TSF2, CFStrust and AFStrust models. This figure obviously shows that, the correct detection
ratio of malicious nodes with trust management models increases sharply with the increase of
simulation time (0~60s), then these values are stable, finally the best one (AFStrust) reaches
0.95. Due to the detailed and complete trust evaluation mechanism (e.g. based on multiple
decision factors) and fuzzy logic rules prediction method in AFStrust model, this model makes
a more accurate detection ratio of malicious nodes and its performance is better than TSF2 and
CFStrust models. Comparing with CFStrust, at the time 120s, the correct detection ratio of
malicious nodes is increased by 3%.
2) Prevent from cheating attacks
In order to cheat a high reputation, a malicious node may make good performance (or disguise
itself to be a good node) in a special time interval, and then behave badly. Proposed trust
models are not successful in prevent from this attack, and the node’s trust value based on
those models has big volatilities. Based on two novel trust’s decision factors (i.e., detailed
recommendation trust and incentive function) and fuzzy logical rules prediction method, our
trust model has ability to prevent malicious nodes from cheating attacks. The comparison
performance with different trust models is shown in Figure 5.
Figure 5 Trust value fluctuation curve of malicious node
Figure 5 shows that, the effect of attack on our model is less than that on TSF2 and CFStrust,
and the malicious nodes’ trust value with AFStrust is smaller than that with TSF2 and
CFStrust. At time 0s, an unknown (malicious) node’s trust value is initialized to 0.7. With the
increase of the simulation time, this malicious node is found by the system respectively based
on different trust models, and the AFStrust has a better performance than TSF2 and CFStrust
in malicious node early recognition. In order to cheat a high trust value, a malicious node
makes good performance in time interval (120s~150s). From Figure 5, we easily see that
AFStrust model effectively mitigate the harm caused by such attacks, while this node’s trust
values based on TSF2 and CFStrust have big volatilities. Due to the special decision factor
(i.e., incentive function), fuzzy logic rules and black-list mechanism in our trust model, in the
middle and later periods, this malicious node’s trust values based on AFStrust decrease gently
while that with the other two models maintain a regular value. This experiment also hints that
the trust obtains hardly while it loses easily.
6. CONCLUSIONS AND FUTURE WORKS
This article studies the questions related with the definition and synthesis of nodes’ trust value
for MANET. Basing on the inherent characteristics of this network, a subjective trust
management model for MANET (AFStrust) based on the analytic hierarchy process theory
and fuzzy logic rules prediction method is proposed. Building weight mechanism based on the
AHP theory to calculate the weights of multiple trust decision factors and handle fuzzy logic

Rao
rules prediction method to predict the node’s trust value make our model more stable,
adaptive and robust, which consequently enhances network’s security and performance. The
simulation results analyze the effectiveness of our trust management model. This model
works as an intuitive and effective evaluation, analysis and derivation tool, it could provide
effective support for the trust decisions and against attacks.
We would continue our work in the following three directions: 1. Make a further
improvement for the trust model proposed in this paper, we plan to incorporate other decision
factors to this model; 2. We will consider an adaptive trust level classification of nodes taking
into account the average trust value of all nodes. The problem of dynamic behavior modification
will also be considered; Moreover, 3. Propose a detailed trust-based on-demand multi-path
routing for MANET in which the node’s trust value is calculated by our trust model.
REFERENCES
[1] Yan Lindsay Sun, Zhu Han, Wei Yu, and K J.Ray Liu.: ‘Attacks on trust valuation in
distributed networks’, Proc. of the 40th Annual Conference on Information Sciences and
Systems, March 2006, pp. 1461-1466.
[2] A.A. Pirzada, and C. McDonald.: ‘Trust establishment in pure ad-hoc networks’, Wireless
Personal Communications, 2006, 37, pp. 39-168.
[3] L. Mui.: ‘Computational models of trust and reputation: agents, evolutionary games, and
social networks’, Ph.D. Thesis. MIT. Massachusetts, 2003.
[4] M.H. Durad, Yuanda Cao, Liehuang Zhu.: ‘Two novel trust evaluation algorithm’, Proc. of
the Communications Circuits and Systems, June 2006, Vol.3, pp. 1641-1646.
[5] George.: ‘Distributed trust evaluation in ad hoc networks’, Proc. of the 3rd ACM workshop
in Wireless Security, Oct. 2004, pp.1-10.
[6] Junhai Luo, Mingyu Fan.: ‘A subjectiveive trust management model based on certainty-
factor for MANETs’, Chinese Journal of Computer Research and Development, 2010,
47(3), pp. 515-523.
[7] Hui Xia, Zhiping Jia, Xin Li, Feng Zhang.: ‘A Subjectiveive Trust Management Model
based on AHP for MANET’, Accessed by the 2011 International Conference on Network
Computing and Information Security (NCIS’11).
[8] Xiaoyong Li, Xiaolin Gui.: ‘Trust Quantitative Model with Multiple Decision Factors in
Trusted Network’, Chinese Journal of Computers, 32, 2009, pp. 405-416.
[9] T.L. Satty.: ‘The analytic hierarchy process’, New York: McGraw-Hill, 1980.
[10] Timothy J. Ross.: ‘Fuzzy Logic with Engineering Applications’, McGraw Hill International
Editions, International Editions, 2000.
[11] U. Wilensky, Netlogo, 1999, http://ccl. northwestern. edu/netlogo/.
[12] Liangchen Wen, Xuefeng Zhang, Limei Zhu.: ‘Method of ameliorative multi-objective
synthetic evaluation based on entropy weight and its application’, Control and Decision
Conference, 2009. CCDC '09. Chinese, pp. 1538-1541.

AN INDEPENDENT TRUST MODEL FOR MANET BASED ON FUZZY LOGIC RULES

More Related Content

What's hot (15)

Similar to AN INDEPENDENT TRUST MODEL FOR MANET BASED ON FUZZY LOGIC RULES (20)

More from IAEME Publication (20)

Recently uploaded (20)

AN INDEPENDENT TRUST MODEL FOR MANET BASED ON FUZZY LOGIC RULES