Under Review
1
SAFE: Spatially-Aware Feedback Enhancement for Fault-Tolerant Trust Management in VANETs
arXiv:2604.07552v1 [cs.NI] 8 Apr 2026
İpek Abasıkeleş Turgut Computer Engineering Department, Iskenderun Technical University Hatay, 31200, Türkiye. E-mail: [email protected] Abstract: Trust management in VANETs is critically important for secure communication between vehicles. In event-based trust systems, vehicles broadcast the events they witness to their surroundings and send feedback reports about other vehicles to a central authority. However, when the event status changes, vehicles that have left the witness area cannot see this change and produce erroneous feedback. This leads to unfair penalization of honest nodes. To solve this problem, the SAFE (Spatially-Aware Feedback Enhancement) approach is proposed. In SAFE, vehicles continue to record messages as long as they remain in the witness area and send updated feedback reports before leaving the area. Additionally, by keeping records between witness and decision distances, more accurate evaluation is ensured. SAFE and TCEMD were compared in singleevent, multi-event, and different decision distance scenarios. The results clearly demonstrate SAFE’s superiority. In single-event, feedback report count increased 2.5 times, and in multi-event, it increased over 6 times. Negative feedback rate dropped from 77% to below 1%. While TCEMD incorrectly blacklisted 34 nodes, this number remained at 1 in SAFE. Even when the decision distance was reduced to 200 m, SAFE showed high accuracy. The findings show that SAFE protects honest nodes in attack-free systems and increases network reliability. Keywords: VANET; trust management; event-based systems; feedback optimization; fault tolerance; vehicular networks; spatial awareness
1 Introduction In smart urbanization, which is a result of globalization and technological developments, autonomous vehicles that can communicate wirelessly with each other have an important place. VANET (Vehicular Ad-hoc Network), one of the fundamental components of intelligent transportation systems, improves traffic flow and driving experience by enabling communication between vehicles (V2V) and with roadside units (V2I). Thanks to the rapid dissemination of emergency messages in the network, both accident prevention and reduction of traffic congestion become possible [1]. However, when the rapidly changing topology of these networks due to high mobility is combined with the unreliability of the wireless environment, it makes the network vulnerable to various attacks. Providing reliable communication is critically important, especially in emergency and safety applications. Trust-based approaches, which offer both low complexity and low computational cost, have therefore found wide application in VANETs [2]. Copyright © —-
2 In event-based trust evaluation systems for VANETs [3–9], vehicles that witness a change in an event’s status broadcast warning messages about the event to other vehicles within their coverage area. Vehicles receiving these messages make a one-time decision based on both their distance to the event and the reliability of the incoming messages and the message sender. Two factors are decisive in this decision: first, the vehicle’s distance to the event, and second, the trustworthiness of the warning sender. In VANET trust systems, to reduce false positive rates, vehicle trustworthiness is typically calculated periodically by a central authority (CDU) and disseminated to the network. In the CDU’s evaluation, personal opinions reported by vehicles (feedback reports) are used. When a vehicle witnesses an event, it evaluates the warning messages it previously received about this event, assigns a positive or negative score to the reporting vehicle, and transmits this score to the CDU. In an ideal system where all vehicles are honest, both the warnings broadcast by vehicles witnessing an event and the reports vehicles provide about each other are expected to be consistent with reality. However, a critical problem is overlooked in existing approaches: when the event status changes, erroneous feedback reports are generated because vehicles that have left the witness area do not have up-to-date information. The timing at which vehicles make their evaluations can cause results inconsistent with reality in their reports. If a vehicle evaluates too early, it will not be able to evaluate new reports that may come in case of a possible event status change. This situation will lead to both making wrong decisions and making incorrect evaluations about other vehicles. On the other hand, while a vehicle’s delay in decision-making allows for more accurate evaluation of other vehicles, it may prevent taking precautions against the event. Therefore, the timing of this decision is critically important in terms of the balance (trade-off) between taking action and making the right decision. Recording messages based on distance and action planning was first proposed by TCEMD [5] and later used in other trust models [6]. However, these studies focused on trust updates in networks under attack and did not examine fault tolerance situations caused by distance in attack-free systems. It has been observed that no study has been conducted on distance-related wrong decision making and optimization of feedback counts. To address these shortcomings, the SAFE (Spatially-Aware Feedback Enhancement) approach is proposed. To compare the performance of SAFE and TCEMD, events of different types and numbers were created in the simulation environment. In the absence of attacks, the expectation is that all nodes provide positive reports about each other and that nodes’ global trust values remain above the threshold value. To evaluate these results, feedback report counts, negative feedback rates, blacklist rates, and nodes’ global trust values were measured. Additionally, to observe how the decision distance (Dd ) affects the decision, system performance was evaluated at different Dd values. The results obtained showed that the SAFE method increased feedback report count up to 6 times compared to TCEMD, reduced the negative feedback rate from 77% to below 1%, and decreased the unfair blacklist count from 34 nodes to only 1 node. These findings demonstrate that the proposed approach prevents unfair penalization of honest nodes and significantly increases network reliability. The main contributions of this study can be summarized as follows:
• A realistic event model for event-based trust systems is proposed and evaluated with parameters at different severity levels (Type 1/2/3).
SAFE: Spatially-Aware Feedback Enhancement for Fault-Tolerant Trust Management in VANETs3 • By extending the distance-based action plans in the literature, vehicles continue to record data as long as they remain in the witness area and send updated feedback reports before leaving the witness area. • The effect of decision distance (Dd ) on system performance is examined and the optimal Dd ≥ 2 × Dw relationship is determined. • By optimizing the feedback report count, the central authority is enabled to make more comprehensive evaluations. The remainder of this paper is organized as follows: Section 2 details the system framework, VANET architecture, event model, and action plans of the proposed SAFE approach. Section 3 introduces the simulation environment and presents the results obtained in single-event, multi-event, and decision distance scenarios. Finally, Section 4 summarizes the conclusions of the study and provides recommendations for future work.
2 System Framework In the modeled VANET system (Figure 1), there are 3 types of elements: vehicles, road-side units (RSU), and Central Decision Unit (CDU). RSUs are used only as a transfer medium between different communication technologies in the communication between vehicles and CDU. Vehicles broadcast the status of the event they witness to their surroundings in an event message (EM). Additionally, they prepare feedback reports (FB) for the relevant vehicles based on the warning messages they previously received about that event. The CDU periodically calculates/updates global trust values based on the reports from vehicles and uploads them to RSUs to be transmitted to vehicles. System assumptions are listed below in accordance with the literature. • Vehicles are equipped with IEEE 802.11p OBU modules that enable communication with other vehicles and RSUs [5, 10–13]. • Vehicles calculate local trust values independently and transmit these values when they detect an RSU within their coverage area [10, 12]. • Since there is no authority governing vehicles, they can be either trustworthy or malicious. • RSUs transmit the local trust values collected from vehicles via wireless communication to the CDU through the backbone network [5, 10–12]. • It is assumed that there are enough RSUs along the route to collect and transmit trust values completely. That is, placement is done in such a way that a vehicle will always be successful when it needs access to an RSU [5, 10]. • The CDU is where global trust is calculated. It is assumed to have sufficient storage and computational resources [5, 10–12]. • The CDU and RSUs are completely trustworthy [5, 10, 11]. • Vehicles joining the network must register with the CDU [11]. • The time of all elements in the network is synchronized [11]. • Vehicles are equipped with modules such as GPS or BeiDou system (BDS) to accurately know the location of themselves and all other nodes [5, 7].
4
Figure 1 An event-based VANET architecture
• For a vehicle to be blacklisted, its threshold value must be below 0.2. Feedback reports sent by a blacklisted node are not considered by the CDU in trust calculations; likewise, event messages broadcast by this node are not evaluated by other vehicles in the decision process. • The initial value is set to 0.5 to be neutral. Vehicles above this value are assumed to be honest. Vehicles that have not yet been blacklisted but are not categorized as honest are classified as suspicious. Although event-based trust systems have been proposed in the literature [3–9], there is no detailed information about how events are modeled, event diversity, or their effects. In previous work [14] of SAFE, an "event" class was first created to manage different events in traffic. The attributes of the event class are shown in Figure 2. Each event was started at a random time and at a random location on the vehicles’ routes in the simulation. The duration, effective distance, and severity of the event vary according to the event type. Events have two different end times: stop time and lasting time. By default, the event status is assumed to be 0 (non-existent). After the event occurs (i.e., Tstart ), that is, after it exists (1), its status continues to be 1 for a duration. Then (i.e., Tstop ) it becomes 0 again. Vehicles witnessing the event broadcast both when the status is 1 and when it becomes 0 again. However, lasting time (i.e., Tlasting ) was added to prevent broadcasting status messages for an event with status 0 after a while and thus prevent broadcast storm formation. At the end of the lasting time, the event status is not broadcast. The response of vehicles throughout the lifetime of an event can be seen in Figure 3.
SAFE: Spatially-Aware Feedback Enhancement for Fault-Tolerant Trust Management in VANETs5 Effective Distance Lasting Time
Start Time Stop Time
EVENT
Type
Severity Duration
Location
Figure 2 The attributes of class “Event”
event = 1 broadcast
start
0
t > Tstart
t > Tlasting
1 t > Tstop
1_0 event = 0 broadcast
Figure 3 AThe response of vehicles throughout the lifetime of an event
Table 1 Characteristics of modeled events
Parameters
Type 1
Type 2
Type 3
Severity
Low
Medium
High
Effective Distance
100 meters
400 meters
800 meters
Duration*
10 min
45 min
2 hours
*The duration parameter reflects the expected values in real traffic environments. Shorter durations were used in the simulation study to obtain distinguishable results
Events modeled in this study are divided into 3 categories according to their severity levels. Category 1 covers events with the lowest severity level, short effective distance, and short duration. In real traffic environments, events such as traffic light warnings and available parking space alerts fall into this category. Category 2 includes events with medium-level severity, with a wider effective area and duration compared to Category 1. Accidents on side roads and congestion on the route fall into this category. The last category includes events with the largest scale and highest severity level. Accidents on main roads and disaster situations fall into this category. The parameter values for each event category are shown in Table 1. The behavior patterns of vehicles according to distances are presented in Figure 4. As seen in Figure 4(a), vehicles within the effective distance of the event (i.e., Dw ) are the
6
Distance < Dw
Dw
Dd
Event
Dw – Dd = Dd
Di
Dd – Di > Di (a)
Action ➢ Broadcast event state ➢ Send feedback report ➢ Record broadcasts ➢ Send feedback report once if needed ➢ Record broadcasts ➢ Decide once and take appropriate action ➢ Record broadcasts ➢ Ignore (b)
Figure 4 (a) Behavior modeling distances for event-related vehicles and (b) their behaviors according to these distances
vehicles witnessing the event. A vehicle must make a decision before entering this distance. That is, the decision distance (i.e., Dd ) is expected to be greater than the effective distance (i.e., Dd > Dw ). In addition to these distances, TCEMD defined a distance where the vehicle is not yet interested in the event, does not evaluate incoming messages related to the event, and concludes that it is far from the event. Vehicles beyond this distance neither evaluate the event nor pay attention to incoming messages. If this distance is called distance of interest (i.e., Di ), the result Dw < Dd < Di will be derived. In this study, first the effect of Dd distance on the decision made in an attack-free system was examined. Then, as shown in Figure 4(b), action plans were designed to ensure that vehicles provide correct feedback reports. In the proposed approach, unlike the literature, a vehicle continues to evaluate incoming event messages as long as it remains in the witness area. For this purpose, it stores broadcasts related to the event and sends a feedback report based on incoming messages once before leaving the witness area to prevent broadcast storm. Another optimization is that the vehicle continues to record event messages between Dw – Dd distances. Although these records do not affect the decision made about the event, they will affect the feedback reports the vehicle will provide and therefore will affect the trust values of both itself and other nodes. These records are accumulated and taken into evaluation when the vehicle witnesses the event. The proposed method will be effective in trust calculations while not changing the decisions of vehicles. In an ideal system, all vehicles are expected to give positive feedback (score = 1) about each other. However, errors may occur due to information delays during the event status change process. For example, let the event status be 1 (active) while a vehicle vi is in the witness area of the event (De < Dw ). The vehicle will honestly broadcast this status to its surroundings. However, when vi moves and leaves the witness area, it cannot witness the change in event status. If the event status changes to 0 (passive) during this time, other vehicles observing the new status will give negative feedback (score = -1) about vi . This situation is the main source of error leading to unfair penalization of honest nodes. The proposed SAFE approach solves this problem by having vehicles continue to record data as
SAFE: Spatially-Aware Feedback Enhancement for Fault-Tolerant Trust Management in VANETs7 initialize
event detected
RSU detected
EM arrived
registered
De < Dw
De < Dd record message
Register RSU warn neighbours create FB report
has FB to sent
send FB to RSU clear FB when ACK arrived
warn neighbours send FB to RSU keep recording send FB once more (if needed)
De > Di
make decision warn neighbours if trusted
ignore message
Figure 5 Vehicle flowchart throughout lifetime
long as they remain in the witness area and send updated feedback reports before leaving the witness area. During the movement period, vehicles perform 3 types of checks as shown in Figure 5: detecting surrounding events, detecting entry into RSU coverage area, and evaluating EMs from other vehicles. When an EM arrives, the vehicle checks the distance (i.e., De ) between the event’s location and itself. If this distance is greater than Di , it ignores the message. Otherwise, it records the message and the event’s location. When the distance to the event falls below Dd , it makes a decision once. After the decision is made, it continues to record event messages so that it can create a more up-to-date feedback report when it witnesses the event. When the vehicle witnesses the event, it broadcasts the actual status of the event and sends the feedback report to the RSU. The vehicle continues to record event reports as long as it remains in the event witness area. When leaving the witness area, it creates one more feedback report if necessary and sends it to the RSU.
3 Results and Discussion In this section, the performance of the proposed approach (SAFE) was analyzed using Omnet++ [15], Veins [16], and SUMO [17] simulators with TCEMD, which forms the basis of event-based trust mechanisms in the literature. Evaluations were conducted based on the metrics of decision accuracy, node classification success, stability of trust values over time, and reliability of feedback reports. To quantitatively evaluate the performance of the proposed SAFE model, the following metrics are defined: 1. The number of feedback reports (FBR): Refers to the total number of feedback reports that vehicles transmit to the CDU through RSUs about the events they witness. This metric is used to measure the density of information accumulation in the network. The CDU will use these reports during global trust calculations and then delete them. In each global trust calculation, feedback reports collected between two consecutive calculations will be used.
8 2. The number of updated global trusts (UGT): Refers to the number of unique nodes whose global trust values are recalculated based on incoming new feedback reports in each period. In global trust calculation, the nodes whose global trust values will be calculated/updated will be determined according to incoming feedback reports. Therefore, both the number and identities of updated nodes will be different in each calculation period. 3. Positive/negative feedback rate: Represents the ratio of honest (score = 1) and erroneous/negative (score = -1) notifications among the reports reaching the CDU. These rates are calculated using (1) and (2). P Positive FBR rate =
FBR, if score = 1 P FBR
P Negative FBR rate =
FBR, if score = −1 P FBR
(1)
(2)
A low Negative FBR Rate represents the system’s resistance to physical errors (fault tolerance). 4. Blacklist/Non-Blacklist rate: As is seen in (3) and (4), the ratio of vehicles whose trust values fall below the determined threshold value (T hB ) as a result of global trust calculation to the total number of evaluated vehicles. It measures the system’s tendency for false penalization (false positive). P Blacklist rate (BLR) =
Vi , if GT (Vi ) ∈ U GT & GT (Vi ) ≤ T hB P Vi , if GT (Vi ) ∈ U GT P
Non-Blacklist rate (N_BLR) =
Vi , if GT (Vi ) ∈ U GT & GT (Vi ) > T hB P Vi , if GT (Vi ) ∈ U GT
(3)
(4)
3.1 Effect of the designed action plan In this section, the effects of the proposed SAFE approach and extended action plans on system performance are examined.
3.1.1 Single event scenario The parameters related to the modeled system are shown in Table 2. In this model, a Type 1 event was created at position (x = 2000 m, y = 400 m) on a 3-lane, one-way highway, 50 seconds after the simulation started. The event remained active for 150 seconds and then became passive. After the 275th second, broadcast messages regarding the presence or absence of the event were stopped. The event witness distance (Dw ) was set to 100 meters, decision distance (Dd ) to 300 meters, and interest distance (Di ) to 500 meters. Vehicles can communicate with both other vehicles and RSUs using broadcast method within a 300meter coverage area. The CDU performs global trust updates every 60 seconds if it has sufficient feedback reports. The initial trust value of nodes was assigned as 0.5, considered as neutral. If a node’s global trust value falls to 0.2 or below, the node is blacklisted and removed from the network. Feedback reports sent by a blacklisted node are not considered by the CDU in trust
SAFE: Spatially-Aware Feedback Enhancement for Fault-Tolerant Trust Management in VANETs9 Table 2 Network parameters and values for single-event scenario
Parameters
Value
Event type
1
Tstart
50 s
Duration
150 s
Tstop
200 s
Tlasting
275 s
Dw
100 m
Dd
300 m
Di
500 m
Global trust update frequency
60 s
Blacklist threshold trust (T hB )
0.2
Initial trust (InTr)
0.5
Event location
(2000m,400m)
Traffic model
3 lane single direction highway
Road distance
8000 m
Coverage radius of vehicles
300 m
calculations; likewise, event messages broadcast by this node are not evaluated by other vehicles in the decision process. Figure 6 shows the number of feedback reports (FBR) evaluated at the times when global trust updates were made for TCEMD and SAFE. Global trust calculation is performed periodically every 60 seconds. Since the event became active at the 50th second, sufficient feedback reports could not be collected by the CDU in the first GT calculation (t = 60 s) and no node’s GT value changed. The first update occurred at the 120th second, which is the 2nd evaluation period. As shown in the figure, the number of feedback collected in all GT periods in the proposed approach is higher than TCEMD. At the 120th second, TCEMD produced 44 reports while SAFE reached 108 reports. This difference became more pronounced at the 180th second, with SAFE producing 153 reports compared to TCEMD’s 60 reports. Similarly, at the 240th second, TCEMD collected 65 and SAFE collected 146 reports; at the 300th second, TCEMD collected 18 and SAFE collected 72 reports. These results show that the SAFE method produces on average 2.5 times more feedback reports. The high number of feedback means that the CDU has more comprehensive data in trust calculation. The fact that the SAFE method continues to record data as long as the vehicle remains in the witness area has significantly increased the data pool available to the center, allowing for more accurate decisions. Since no attack is modeled in the designed system, the trust values of nodes change only based on their broadcasts and feedback reports. In an ideal situation, no node is expected to give a negative score about another. However, errors may occur due to information delays during the event status change process.
10
The number of feedback reports (FBR)
180 160 140 120 100 80 60 40 20 0
SAFE TCEMD
120 108 44
180 153 60
240 146 65
300 72 18
Global Trust Estimation Time (sec) Figure 6 The number of feedback reports of SAFE and TCEMD in each global trust update period for single-event scenario
For example, let the event status be 1 (active) while a node vi is in the witness area of the event. The node will honestly broadcast this status to its surroundings. However, when vi moves and leaves the witness area, it cannot witness the change in event status. If the event status changes to 0 (passive) during this time, other nodes observing the status will give negative feedback about vi . This situation is the main source of error leading to unfair penalization of honest nodes. In the proposed approach, action plans were designed as explained in Section 2 to reduce the negative feedback rate. Figure 7 shows the positive and negative feedback rates in both TCEMD and SAFE methods. It should be noted that no negative feedback occurs until the event status changes from 1 to 0. When the event status changes (Tstop = 200 s, i.e., before the 4th GT calculation), negative feedback emerges. The figure shows the rates for the 4th and 5th GT periods. The negative feedback rates, which were 52% and 56% respectively in TCEMD, were reduced to 25% and 8% in the proposed method. This reduction has ensured more accurate calculation of global trust values of honest nodes. The UGT values and blacklist rates measured for all update periods are shown in Figure 8 and Figure 9, respectively. As seen in Figure 8, the number of unique nodes whose GT values were updated (UGT) in the proposed method remained above TCEMD in all time periods. For example, at the 180th second, SAFE updated 32 nodes while TCEMD could only update 25 nodes. This difference shows that the SAFE method can evaluate a wider range of vehicles and increases the overall reliability visibility of the network. When the blacklist rates are examined in Figure 9, the blacklist rate reaching 67% in TCEMD in the 4th GT period (t = 240 s) when the event status changed was reduced to 21% in the proposed approach. In the 5th GT period (t = 300 s), this rate decreased from 45% to 13%. This improvement was achieved thanks to the extended action plan enabling vehicles to collect up-to-date information before leaving the witness area. Thus, even if
5th
TCEMD
5th
SAFE
4th
TCEMD
4th
SAFE: Spatially-Aware Feedback Enhancement for Fault-Tolerant Trust Management in VANETs11
SAFE
44
56
92
8
48
52
75 0
20
40
Positive FBR
25 60
80
100
Negative FBR
The number of updated global trusts (UGT)
Figure 7 Positive and negative FBR rates of SAFE and TCEMD at 4th and 5th global trust update time for single-event scenario
35 30 25 20 15 10 5 0
SAFE TCEMD
120 24 18
180 32 25
240 28 24
300 16 11
Global Trust Estimation Time (sec) Figure 8 The number of updated global trusts of SAFE and TCEMD in each global trust update period for single-event scenario
5th
TCEMD
5th
SAFE
4th
TCEMD
4th
12
SAFE
21
79
13
87
33
67
45 0
20
55 40
N_BLR
60
80
100
BLR
Figure 9 Blacklist and Non-Blacklist rates of SAFE and TCEMD at 4th and 5th global trust update time for single-event scenario
the event status changes, vehicles can produce more accurate reports about each other and honest nodes are not unfairly penalized. The results obtained demonstrate that the proposed SAFE method provides a significant superiority over TCEMD in the single event scenario. SAFE increased the number of feedback reports by an average of 2.5 times, allowing the center to make more comprehensive evaluations. The negative feedback rate was reduced from 52% to 25% and from 56% to 8%; the blacklist rate was reduced from 67% to 21% and from 45% to 13%, significantly preventing unfair penalization of honest nodes. This success was made possible by the vehicle continuing to record data as long as it remains in the witness area and sending an updated feedback report before leaving the witness area.
3.1.2 Multi event scenario In this scenario, 3 different types of events were created on the vehicles’ routes. The characteristics of event types are detailed in the System Framework section; their parameters are shown in Table 3. Type 1 represents the smallest scale event (Dw = 100 m), Type 2 represents the medium scale event (Dw = 400 m), and Type 3 represents the event with the widest effective area (Dw = 800 m). Each event has a different start time (60 s, 250 s, 150 s), duration (100 s, 50 s, 100 s), and location. Thanks to this diversity, the performance of the methods against events of different scales can be evaluated. To compare the performance of SAFE and TCEMD, FBR values according to GT calculation time are shown in Figure 10, and positive and negative FBR rates in the 4th and 5th GT periods are shown in Figure 11. As seen in Figure 10, SAFE collected significantly more feedback reports compared to TCEMD in the multi-event scenario. At the 120th second, SAFE produced 100 and TCEMD produced 42 reports; at the 180th second, this difference became pronounced with SAFE’s 465 reports compared to TCEMD’s 78 reports. The highest difference was observed at the 300th second, where SAFE collected 882 and TCEMD could only collect 134 reports (a 6.6-fold difference). Compared to the average 2.5-fold increase in the single event scenario, this ratio exceeded 6 times in the multi-event scenario.
SAFE: Spatially-Aware Feedback Enhancement for Fault-Tolerant Trust Management in VANETs13
1000 900
800 700
FBR
600 500 400 300 200 100 0 SAFE TCEMD
120 100 42
180 465 78
240 343 142
300 882 134
Global Trust Estimation Time (sec)
5th
TCEMD
5th
SAFE
4th
TCEMD
4th
Figure 10 The number of feedback reports of SAFE and TCEMD in each global trust update period for multi-event scenario
SAFE
22,4
77,6
99,1
0,9
66,4
33,6
99,5 0
20
40
Positive FBR
0,5 60
80
100
Negative FBR
Figure 11 Positive and negative FBR rates of SAFE and TCEMD at 4th and 5th global trust update time for multi-event scenario
14
Event Parameter
Type 1
Type 2
Type 3
ID
1
2
3
Tstart
60 s
250 s
150 s
Duration
100 s
50 s
100 s
Dw
100 m
400 m
800 m
Dd
200 m
600 m
1000 m
Di
400 m
800 m
1200 m
Location
(1000m, 400m)
(3000m, 400m)
(2000m, 400m)
60
30
50
25
40
20
TCEMD _ UGT
SAFE _ UGT
Table 3 Network parameters and values for multi-event scenario
30 20
10 0 Event_1 Event_2 Event_3
120 21
180 29
240 20
41
33
300 45 52
360 28 23
Global Trust Estimation Time (sec)
(a)
15 10
5 0
Event_1 Event_2 Event_3
120 17
180 23
240 15
10
28
300
360
18 24
14 1
Global Trust Estimation Time (sec)
(b)
Figure 12 Event-based updated global trusts of (a) SAFE and (b) TCEMD in each global trust update period for multi-event scenario
Figure 11 examines the positive/negative FBR rates in the 4th and 5th GT periods. While the negative FBR rate in the SAFE method remained below 1% in both periods (0.5% and 0.9%), this rate increased to 33.6% in the 4th period and 77.6% in the 5th period in the TCEMD method. These results show that SAFE minimizes erroneous negative notifications in the multi-event scenario as well. To analyze the performance of TCEMD and SAFE on an event basis, the number of nodes whose GT values were updated (UGT) was measured for each event. The results are shown in Figure 12. Figure 12(a) shows SAFE results and Figure 12(b) shows TCEMD results. UGT values for all events in the SAFE method remained above TCEMD. Especially for Event 3 (Type 3), which has the widest effective area, SAFE updated 41 nodes at the 180th second while TCEMD could only update 10 nodes. Similarly, at the 300th second, SAFE updated 52 and TCEMD updated 24 nodes. This difference shows that the SAFE method can evaluate a wider range of vehicles and increases the overall reliability visibility of the network. In
SAFE: Spatially-Aware Feedback Enhancement for Fault-Tolerant Trust Management in VANETs15 Table 4 Trust classification of nodes for each event
Event ID
Method
Untrusted
Suspicious
Honest
1
TCEMD SAFE
6 1
1 1
48 68
2
TCEMD SAFE
11 0
0 1
21 72
3
TCEMD SAFE
17 0
2 0
44 149
addition, nodes were labeled in three categories according to their GT values: untrusted (blacklist), suspicious, and honest. The label of a vehicle vi depends on the value it receives after the global trust measurement. If the node’s GT value is below the blacklist threshold (T hB = 0.2), it is labeled as untrusted ; if below the initial trust value (InT r = 0.5), it is labeled as suspicious ; otherwise, it is labeled as honest. The classification distribution of nodes after each event is shown in Table 4. The labeling criteria used to measure the success of the proposed approach are formulated in (5). U ntrusted, if GT (Vi ) ≤ T hB (5) LabelVi = Suspicious, if T hB < GT (Vi ) ≤ InT r Honest, otherwise When the data in Table 4 is examined, it can be seen that the SAFE method provides a significant superiority over TCEMD. In the TCEMD method, honest nodes were unfairly penalized due to timing errors caused by distance. Especially in the Type 3 event, which has the widest effective area, 17 nodes were incorrectly labeled as untrusted. In contrast, the SAFE method reduced the number of untrusted from 6 to 1 in the Type 1 event; and completely zeroed this number for Type 2 and Type 3 events. In total, while TCEMD unfairly blacklisted 34 nodes, SAFE reduced this number to only 1. In the suspicious category, while TCEMD labeled a total of 3 nodes as suspicious, this number remained at 2 in the SAFE method. Additionally, the SAFE method significantly increased the number of nodes labeled as honest in all three events: from 48 to 68 in Type 1, from 21 to 72 in Type 2, and from 44 to 149 in Type 3. This increase in Type 3, which has the widest effective area (3.4-fold), is particularly noteworthy. These results demonstrate that the extended action plans prevent unfair penalization (false positive) of honest nodes in multi-event scenarios and increase the overall reliability of the network. When the multi-event logs are examined in detail, it was observed that the SAFE method not only kept honest nodes in the system but also rapidly increased their trust values. For Type 3, which has the widest effective area, at the 180th second only 4 nodes could exceed the initial trust value (InT r = 0.5) in the TCEMD method, while this number reached 41 in the SAFE method. More importantly, by the 300th second in the TCEMD method, 11 nodes were unfairly blacklisted due to erroneous negative reports, while this number remained at zero in the SAFE method. This situation is visualized in Figure 13. Figure 13(a) shows the change over time in the number of nodes blacklisted for the Type 3 event. While significant increases were observed at t = 180 and t = 300 (6 and 11 nodes respectively) in the TCEMD method, the SAFE method remained stable with zero blacklist in all time periods. Figure 13(b) shows the change in the number of honest nodes. While the SAFE
16 60
Number of Honest Nodes
Number of Blacklisted Nodes
12 10
8 6 4 2 0 SAFE TCEMD
120 0 0
180 0 6
240 0 0
300 0 11
360 0 0
Global Trust Estimation Time (sec)
(a)
50
40 30 20 10 0
SAFE TCEMD
120 0 0
180 41 4
240 33 28
300 52 11
360 23 1
Global Trust Estimation Time (sec)
(b)
Figure 13 (a) The number of nodes blacklisted and (b) determined as honest in each global trust update period for Event-3 in multi-event scenario
method labeled 41 nodes as honest at t = 180 and 52 nodes at t = 300, TCEMD could only reach 4 and 11 nodes respectively. These results show that TCEMD makes sudden and erroneous decisions during event status changes, while the SAFE method performs consistent and accurate classification. The results obtained in the multi-event scenario demonstrate that the SAFE method provides a significant superiority over TCEMD in all metrics. SAFE increased the number of feedback reports beyond the 2.5 times increase in the single event scenario, achieving an increase of over 6 times. While the negative feedback rate reached up to 77.6% in TCEMD, this rate remained below 1% in the SAFE method. From a node classification perspective, while TCEMD unfairly labeled a total of 34 nodes as untrusted, SAFE reduced this number to only 1. The number of honest nodes significantly increased in all three events; particularly rising from 44 to 149 in Type 3. These results confirm that the extended action plans work effectively in multi-event scenarios as well and preserve network reliability in dynamic traffic environments.
3.2 Effect of decision distance To examine the effect of decision distance (Dd ) on nodes’ feedback reports and global trust calculation, the Dd = 200 m value was also tested in addition to the Dd = 300 m value used in the previous scenario. There are 2 main constraints in the selection of decision distance. First, the Dd value must not exceed the coverage area of vehicles (300 m); otherwise, vehicles will not be able to receive messages from other nodes when they reach the decision point. Second, setting the Dd value too low will result in vehicles not having sufficient time for action planning. Figure 14 shows the FBR values for TCEMD (a) and SAFE (b) at different Dd values. While no significant difference was observed between Dd = 200 m and Dd = 300 m in the TCEMD method, the Dd = 200 m value provided higher FBR especially at t = 300 in the SAFE method (89 vs 72). This result shows that SAFE can prevent data loss thanks to the action plan even at shorter decision distances. Figure 15 details the performance of the SAFE method in the 4th and 5th GT periods for Dd = 200 m. While the negative FBR rate was 11% and the blacklist rate was 7% in the 4th period, both metrics dropped to zero in the 5th period. Considering that TCEMD
SAFE: Spatially-Aware Feedback Enhancement for Fault-Tolerant Trust Management in VANETs17 70
180 160
60
140
SAFE - FBR
TCEMD - FBR
50 40 30 20
Dd = 200m Dd = 300m
100 80 60 40
10 0
120
20 120 45 44
180 64 60
240 64 65
300 40 18
Global Trust Estimation Time (sec)
(a)
0 Dd = 200m Dd = 300m
120 104 108
180 154 153
240 147 146
300 89 72
Global Trust Estimation Time (sec)
(b)
Figure 14 The number of feedback reports of SAFE and TCEMD in each global trust update period for Dd =200 and Dd =300
produced 52%–56% negative feedback rate even at Dd = 300 m, it can be seen that the SAFE method produces much more reliable results even at shorter distances. These findings confirm that the strategy of continuing to record data in the witness area (between Dw –Dd ) successfully manages the decision distance constraint. However, for safe driving and action planning, it is recommended that the Dd distance be kept at least twice the witness distance (Dd ≥ 2 × Dw ). This value provides the balance (trade-off) between the system’s accuracy performance and practical applicability.
4 Conclusion In this study, the SAFE approach is proposed as a solution to the distance-related erroneous feedback problem encountered in event-based trust systems in VANETs. The main principle of SAFE is that vehicles continue to record data as long as they remain in the witness area and send updated feedback reports before leaving the area. The proposed approach was compared with TCEMD in single event, multi-event, and different decision distance scenarios. The results are summarized as follows: • Single Event Scenario: SAFE increased the number of feedback reports by 2.5 times and reduced the negative feedback rate from 52% to 25%. • Multi-event Scenario: The performance gap widened significantly; the number of feedback reports increased more than 6-fold, and the negative feedback rate dropped from 77% to below 1%. • Node Reliability: While TCEMD unfairly blacklisted 34 nodes, this number remained at 1 in SAFE. Furthermore, the number of honest nodes increased from 44 to 149 in Type 3 events. The effect of decision distance (Dd ) was also examined. SAFE demonstrated high accuracy even at Dd = 200m. In the 5th GT period, negative feedback and blacklist rates
18 120
100
%
80 60 40 20 0 4th 5th
Positive FBR 89 100
Negative FBR 11 0
N_BLR 93 100
BLR 7 0
Figure 15 Positive FBR, negative FBR, blacklist rate, and non-blacklist rate percentages obtained for SAFE at the 4th and 5th global trust calculation periods for Dd =200m
dropped to zero, validating the strategy of maintaining records within the witness area even at short decision distances. Nevertheless, Dd ≥ 2 × Dw is recommended for safe driving. In conclusion, SAFE prevents unfair penalization of honest nodes in attack-free systems and increases network reliability. Future work will focus on testing this approach in specific attack scenarios and integrating it with different trust management methods.
Acknowledgement This study was supported by Scientific and Technological Research Council of Turkey (TUBITAK) under the Grant Number 124E017. The author thanks to TUBITAK for their supports.
Declarations Conflicts of Interest: The author declares no conflicts of interest. Use of AI Tools: During the preparation of this manuscript, the author used Claude (Anthropic) for language refinement and proofreading. The author reviewed and edited all AI-assisted content and takes full responsibility for the content of the published article.
References [1] Mengqi Wang, Jiayuan Mao, Wei Zhao, Xinya Han, Mengya Li, Chuanjun Liao, Haomiao Sun, and Kexin Wang. Smart city transportation: A vanet edge computing
SAFE: Spatially-Aware Feedback Enhancement for Fault-Tolerant Trust Management in VANETs19 model to minimize latency and delay utilizing 5g network. Journal of Grid Computing, 22(1):25, 2024. [2] Saeed Shokrollahi and Mahtab Dehghan. Ctvan: a cooperation-based rsu-assisted trust management model for reliable communication in vanets. Cluster Computing, 28(4):227, 2025. [3] Angelo Feraudo, Nicolò Romandini, Carlo Mazzocca, Rebecca Montanari, and Paolo Bellavista. Diva: A did-based reputation system for secure transmission in vanets using iota. Computer Networks, 244:110332, 2024. [4] Mehmet Aslan and Sevil Sen. A dynamic trust management model for vehicular ad hoc networks. Vehicular Communications, 41:100608, 2023. [5] Zhiquan Liu, Jian Weng, Jianfeng Ma, Jingjing Guo, Bingwen Feng, Zhongyuan Jiang, and Kaimin Wei. Tcemd: A trust cascading-based emergency message dissemination model in vanets. IEEE Internet of Things Journal, 7(5):4028–4048, 2019. [6] Jianxiang Qi, Ning Zheng, Ming Xu, Ping Chen, and Wenqiang Li. A hybrid-trustbased emergency message dissemination model for vehicular ad hoc networks. Journal of Information Security and Applications, 81:103699, 2024. [7] Farhan Ahmad, Fatih Kurugollu, Chaker Abdelaziz Kerrache, Sakir Sezer, and Lu Liu. Notrino: A novel hybrid trust management scheme for internet-of-vehicles. IEEE Transactions on Vehicular Technology, 70(9):9244–9257, 2021. [8] Rasha Jamal Atwa, Paola Flocchini, and Amiya Nayak. Rteam: Risk-based trust evaluation advanced model for vanets. IEEE Access, 9:117772–117783, 2021. [9] Arpita Bhargava and Shekhar Verma. Duel: Dempster uncertainty-based enhancedtrust level scheme for vanet. IEEE Transactions on Intelligent Transportation Systems, 23(9):15079–15090, 2022. [10] Jinsong Zhang, Kangfeng Zheng, Dongmei Zhang, and Bo Yan. Aatms: An anti-attack trust management scheme in vanet. IEEe Access, 8:21077–21090, 2020. [11] Jie Cui, Xiaoyu Zhang, Hong Zhong, Zuobin Ying, and Lu Liu. Rsma: Reputation system-based lightweight message authentication framework and protocol for 5genabled vehicular networks. IEEE Internet of Things Journal, 6(4):6417–6428, 2019. [12] Jianxiang Qi, Ning Zheng, Ming Xu, Xiaodong Wang, and Yunzhi Chen. A multidimensional trust model for misbehavior detection in vehicular ad hoc networks. Journal of Information Security and Applications, 76:103528, 2023. [13] Hesham El Sayed, Sherali Zeadally, and Deepak Puthal. Design and evaluation of a novel hierarchical trust assessment approach for vehicular networks. Vehicular Communications, 24:100227, 2020. [14] İpek Abasıkeleş Turgut. The effect of selfish nodes on the performance of eventbased trust management in vanets. In 2025 9th International Symposium on Innovative Approaches in Smart Technologies (ISAS), pages 1–5. IEEE, 2025. [15] Andras Varga. Omnet++. In Modeling and tools for network simulation, pages 35–59. Springer, 2010.
20 [16] Christoph Sommer, David Eckhoff, Alexander Brummer, Dominik S Buse, Florian Hagenauer, Stefan Joerer, and Michele Segata. Veins: The open source vehicular network simulation framework. In Recent advances in network simulation: the OMNeT++ environment and its ecosystem, pages 215–252. Springer, 2019. [17] Michael Behrisch, Laura Bieker, Jakob Erdmann, and Daniel Krajzewicz. Sumo– simulation of urban mobility: an overview. In Proceedings of SIMUL 2011, the third international conference on advances in system simulation. ThinkMind, 2011.