Soft Computing

[Link] (0123456789().,-volV)(0123456789().
,- volV)

FOCUS

Imposition recognition in vehicular networks using hybrid offloading

techniques with improved radial bias neural network
P. Bhanupriya1 • B. Ramachandran1

Accepted: 12 August 2023

 Springer-Verlag GmbH Germany, part of Springer Nature 2023

Abstract
The development of intelligent transportation systems (ITS) has emerged as the most effective and efficient method of
increasing the diversity of existing transportation networks during the past few decades. Because of its important con-
tribution to effectively using network capacity while consuming little energy, vehicular data offloading will be required for
emerging networking technologies like as 5G and D2D. Networking options such as Bluetooth, Wi-Fi and opportunistic
communications are available to send data that would otherwise need the use of a cellular operator’s services. Data
offloading can enhance the usage of network capacity by a large amount. Vehicular data offloading contributes significantly
to the reduction of cellular network bandwidth use. It assists the network in increasing its throughput by allowing many
users to receive the data simultaneously. To define the issue of Vehicular data offloading, we must first determine that it is
an NP-hard target set selection (TSS) problem. TSS is the process of choosing a minimum set of nodes for exchanging
duplicate data by combining Hybrid PSO with GWO to reduce the number of nodes required (Particle Swarm Optimization
with Gray Wolf Optimization). Users and ISPs working together to discover efficient target sets may be able to offer
improvements based on their combined opinions. The existence of malevolent users in the target set might also cause
network activity to be hindered, either by spoofing or by limiting the offloading capability of the network, depending on the
circumstances. Because of these malevolent users, the entire performance of the network may suffer because of their
actions. In this proposed work, we offer a hybrid communication method for defining the set of users who should receive
the message. To overcome the problem of overlapping community detection, we make advantage of the features of opinion
dynamics between users. The target set’s security is handled via the use of trust-based metrics derived from the users’
actions. The proposed work also includes a technique for categorising the chosen offload parameters using radial bias
neural network (RBNN) to reach the emergency number, among other things. Using the suggested work’s performance
measures, it is possible to determine the following: latency, Entropy, and Accuracy.

Keywords NP-hard target set selection (TSS) problem  Offload parameters  Radial bias neural network (RBNN) 
Improved reduced round advanced encryption  Delay  Entropy  Accuracy

1 Introduction

The development of mobile networks is a never-ending

process (Global Networking Trends Report 2020).
According to predictions, the next few years will be crucial
& P. Bhanupriya for network operators and service providers. They must
bp1869@[Link] design network installations with appropriate settings in
B. Ramachandran mind, considering the large number of mobile-enabled
ramachab@[Link] devices (Cisco Annual Internet Report 2020) and the
1
Department of Electronics and Communication Engineering, potential for future applications. At the same time, it is
Faculty of Engineering and Technology, SRM Institute of predicted that data traffic would reach almost 924 Exabytes
Science and Technology, SRM Nagar, Kattankulathur, by the year 2022 (Cisco Visual Networking Index 2019),
Chengalpattu, Chennai, Tamil Nadu 603203, India

123
 P. Bhanupriya, B. Ramachandran

which is a significant increase from the current level. of this. When it comes to more wide networks, it is chal-
Mobile data offloading is a viable method for meeting the lenging to simulate the mobility factor. There is a lot of
demands of growing data traffic. It is becoming more randomness in the network. The randomness in a network
popular. The usage of multimedia apps in day-to-day is not taken into consideration by the graph-based analysis
activities causes an exponential increase in the amount of that is often utilised in network relationships. In realistic
data sent. In recent decades, the communication revolution models, the graphs are always changing. Users or access
has made important contributions to all facets of mankind’s points may only express their opinions on the qualities,
technical growth, particularly in the area of information which are confined to their own opinions. The cumulative
technology. The present developments need the use of influence of communities (Andreev et al. 2014) working
more efficient data offloading solutions, which are together, result in a random distribution of information in
becoming increasingly prevalent. Improved 5G and 6G the network. As a result, users’ opinions on the network
communication systems will provide a wide range of might be used to assess the network. Data offloading
application beneficial for users (Bhanupriya et al. 2022; solutions are protected against unauthorised access. Users
Spatharakis et al. 2020). When consumers have similar in an opportunistic network cannot be relied upon to be
interests, the role of mobile data offloading becomes trustworthy. Malicious nodes are possible, and they have
increasingly substantial. Furthermore, as the data needs are the potential to provide inaccurate or partial information,
more delay tolerant, the impact of this factor increases even raising security problems. To handle safe data forwarding,
more. Furthermore, the deployment of an offloading such individuals must be identified as trustworthy or
method must keep up with the rate of rise in mobile data untrustworthy and dealt with accordingly. Because certain
traffic as a result, such models must overcome the unre- network users are adamant about not sharing their infor-
alistic assumption by using time-dependent analysis while mation with others, even if it can be handled using
taking into consideration their mobility. The latency tol- acceptable network protocols, appropriate off loaders (Han
erance of networks should also be taken into consideration et al. 2010) continue to be used. The lease of bandwidth
(Bhanupriya et al. 2022) throughout this process. The next and four caches is also important for the development of
chapter contains the results of a thorough investigation of suitable reimbursement mechanisms. Because of this, we
the existence of such research gaps. Users’ dynamic must lower the total cost of providing mobile data requests.
behaviour-based features (Bhanupriya et al. 2021 were Using a suitable and secure system for the joint optimiza-
used to try to implement a mobile network with restricted tion of routing rules, it is possible to alleviate this problem
delay tolerance (Gauni et al. 2023), which was the goal of by leasing the resources. Provision of a portion of the
this proposed work. It assists us in developing an effective incentive distribution because of the lack of an incentive,
mobile data offloading strategy that allows just a small the users would feel pressured to engage willingly. There
number of users to act as helpers or off loaders. Our system are just a few incentive schemes in the literature, and most
proposes optimum offloading in opportunistic networks of them are partial, with only a few users being granted the
based on opinion-based trust derivations, which are derived whole authority to decide on the incentive distribution and
from expert opinions. The opportunistic networks are those allocation. Such processes may become biased in their
that form to serve a certain purpose by forming a com- ability to observe the most advantageous incentive
munity around a particular channel of communication schemes. In most schemes, such as the salary-bonus plan,
(Valerio et al. 2014). the foundation makes bidirectional decisions between data
Most of the research uses unrealistic models for devel- needs and compensation, with the foundation acting as the
oping heuristics to identify network relationships, and this intermediary between the two. The reward should not be
is reflected in the results. When it comes to network the same for all off loaders; rather, it should be based on
analysis, the greedy approaches rely on static graphs (Han their level of engagement in the offloading process, which
et al. 2012). In a social network, the graph nodes may should be disclosed beforehand. It is also difficult to sim-
represent users, and the edges may reflect the connections ulate users’ behaviour to address the technological and
between them (Valerio et al. 2014). However, since there economic challenges involved in incentive derivation,
may be several links between distinct qualities, the sub- which are both challenging to do. The issue is addressed by
graphs may overlap, and prioritising may be necessary to certain literature-based models, but they are confined to the
avoid confusion. Even though network heuristics are used mobile network operator (MNO) or the access point (AP),
to assess numerous communities, they are mostly over- and they do not handle it for the consumers.
looked in the literature (Ghorbani et al. 2016). With respect All feasible methods for mobile data offloading in a
to restricted attribute-based connections, greedy techniques variety of opportunistic networks must be identified and
are prone to being partial, with the relevance of most tested. The major aim is to determine if such offloading
characteristics being overlooked in such situations because options are feasible in practical contexts. As the number of

123
Imposition recognition in vehicular networks using hybrid offloading techniques...

services available grows, a variety of obstacles develop in 2 Related survey

the process of adopting these services into the system. The
following are the most significant difficulties that have an A mathematical framework for delay tolerant network
impact on the network’s performance: (DTN)-based mobile traffic offloading is presented by the
Communication time lag (delay in transmission of authors of Valerio et al. (2015). The mathematical frame-
information): as the number of cars in a vehicular network work is based on the concept of delay tolerance. Within the
grows, the communication procedure becomes more time- context of a realistic mobile environment, the model
consuming and difficult. highlights the importance of a variety of various data
• Network failure: As the size of the network grows, the components. Aside from that, they have emphasised the
likelihood of a network failure owing to a malfunc- need of designing efficient offloading mechanisms that take
tioning component of any of the network components into consideration the different demands and interests of
grows as well, as previously noted. Such failures have mobile data users. We can assist in the research of the
the potential to result in the loss of very important study by using real trace-driven simulations. The data
information that is present in the network. demonstrate that the system’s performance in both the
• There is a lack of adequate security: Another issue that human and vehicular environments has been improved
develops because of the expansion of the network is the because of the improvements. One of the study’s conclu-
security of the cars. As the number of cars in the sions was the investigation of offloading amounts for a
network grows, it becomes simpler for an intruder to variety of life lengths and buffer sizes, among other things.
target a single vehicle, which might have far-reaching In Ackerman et al. (2010), authors considered vehicle
consequences for the whole network. interactions with mobile social networking services. When
• Insecure channel: When information is sent across a working in hybrid networks, they make advantage of the
long distance, the information contained inside the help given by access points (APs). This assists in the
vehicle channel has the potential to be intercepted and detection of space crossing communities when space
altered. crossing communities are found in these types of networks.
• Computation time: Another difficulty that occurs before To send information, the researchers have created a
the cars begin communicating in the network is the mechanism of transmission. Similarity attraction and
amount of time necessary to protect the information infrastructure support (SAIS) is a system developed by the
contained in the vehicles. University of Michigan that detects such groupings by
using infrastructure and shared choices. The findings have
In order to keep up with the growing demand for vehicle been compared to the routing algorithms Simbet algorithm
networks, it is necessary to update vehicular networks on a (Liu et al. 2017), and the results show that they are in
regular basis. New algorithms and schemes that are focused accordance with one another. In addition, the Nguyen
on lowering communication time, enhancing fault toler- routing algorithm and the Bubble Rap technique have been
ance, and offering greater security and privacy must be tested in comparison to SAIS to determine their effec-
developed to meet the needs of vehicular networks. This tiveness. Over the course of this inquiry, deliverability
contributes to increasing the efficiency of the network ratios, overhead ratios, and average latency were all dis-
while simultaneously reducing the communication over- covered to be useful indicators. The study is carried out
head. The following are the primary goals of this project, in with the use of data from the MIT and the University of
no order: Illinois. When it comes to data forwarding and community
A robust delay reduction system for vehicle communi- identification, the core principle is to take advantage of the
cation is being developed. collaboration of APs that are physically near together
• To develop a fault-tolerant vehicular network for the (Loukas et al. 2017). When it comes to mobile computing
purpose of facilitating effective vehicle management. settings, offloading is critical, as shown by the research
study reported in Gao et al. (2016). Utilizing the Lyapunov
For the Internet of Vehicles, it is necessary to develop optimization approach, it has been determined how much
security and privacy protection strategies. energy may be saved. As a result, the model provides a
The following is the structure of the proposed paper. In dynamic offloading algorithm (DOA) that is optimised for
Sect. 2, there is a section on related work, in Sect. 3, there energy efficiency. The adaptive dynamic offloading tech-
is a section on the methodology of the proposed study, in nique oversees deciding which software components are to
Sect. 4, there is a section on experimental analysis and be employed in the offloading process. Because it is pos-
results, and in Sect. 5, there is a section on conclusion and sible to conduct operations remotely via wireless networks,
future work. it is advantageous to the components who may then benefit

123
 P. Bhanupriya, B. Ramachandran

from this. They classify components into a few various resources are particularly important for the aim of
categories in order to determine which components should offloading computation-intensive operations. For balancing
be offloaded and which components should not be off- the performance of mobile devices with the consumption of
loaded. A major emphasis of Srinivasan et al. (2006) is on energy during computation search-adjust strategy was
the significance of finding the top limit of user attributes, as employed. A significant portion of the work (Gupta and
well as how to go about establishing this limit. In addition Rohil 2012) is devoted to data items with a clearly defined
to assisting in the identification of users who should be deadline that are reliant on the capacity restrictions of Wi-
designated for content delivery, it also helps to accelerate Fi access points in the surrounding region. A variety of
the dissemination process. It is suggested and executed a offloading options are selected in line with the require-
reinforcement learning technique based on the actor-critic ments of the delivery assurance programme. Even though
method, which is based on the actor-critical method. As the authors acknowledge that users must work together to
part of their approach, the agent must pay attention to the facilitate offloading, they also acknowledge that they must
status of the system, take appropriate action, and then be share resources like as battery power and storage space to
compensated with monetary compensation. The model has do this. In addition to the incentive system, the entire
been evaluated and proved to be accurate using real-world execution is critical to motivate users to finish the pro-
data, according to the results. It has also been evaluated on gramme effectively and successfully. Furthermore, the
synthetic datasets to determine how it compares to previ- need of infrastructure to develop successful offloading
ously reported methods. Bayesian network-based overlap- environments has been emphasised. While it comes to
ping community identification approaches are shown in time-sensitive data items, the researchers at Lee et al.
Hui et al. (2005) via the use of the data supplied in the (2013) take the capacity limitations of access points into
paper. The study illustrates that the user recognised an account when building their solution. It is derived based on
analogous model, which is a significant contribution to the probability theory in order to aid in the assessment of the
vehicle body of knowledge. It is the goal of the study data collected. Verification of the results has, on the other
detailed in Eagle et al. (2007) to integrate device-to-device hand, been accomplished via the use of simulated experi-
(D2D communications with already available cellular ments in synthetic environments. After conducting a thor-
technology standards. The data demonstrates to achieve ough analysis of the literature, as described earlier in this
significant D2D benefits, it is required to develop suit- section, it has become clear that there is a pressing need to
able offloading criteria and to increase the number of precisely define the core elements that will be utilised to
source and destination pairs available. Customers may categorise users in the network. For success, it will help in
experience large packet delays and increased energy usage the modelling of real implementation scenarios as well as
if the QoS standards are not met, which is discussed in the the adoption of a collaborative strategy among consumers
article as a possibility. Additionally, the necessity for and access points of ISPs. When assessing the success of
determining the appropriate D2D network connections is offloading, it is crucial to take into account the influence of
compared to LTE in terms of link lengths, bit rates that are users’ qualities that are constantly changing in the network.
compatible with fair scheduling, and other criteria, among According to the conclusions of a recent literature analysis,
others. They also help to increase the number of high-speed offloading offers a significant contribution to energy con-
connections that are accessible when there is a large vol- servation and conservation (Shahryari et al. 2021). This
ume of traffic on a network via their efforts. The authors example also successfully highlights the requirement for
have also looked at a dynamic model for offloading data both static and dynamic network identification in terms of
traffic, and they have taken this into mind. Using rein- time lengths, which are both necessary. When it comes to
forcement learning to accomplish their aims, the authors in keeping up with the current trend of smartphone usage and
Orimolade and Ventura (2015) have also constructed a industry requirements based on virtual environments, smart
similar kind of opportunistic network that is comparable to sensor deployment or existing infrastructures need to stay
the one described above. In Jiang et al. (2016), the authors on top of both the current trend and the anticipated future
emphasise the use of social media networks to tackle target to keep up with the anticipated expansion of mobile data.
set selection difficulties, with a special focus on the usage The need for effective data dumping systems has increased
of Facebook and Twitter as tools for this purpose. The as a result of the recent global pandemic-like events that
minimal target set and maximum active set approxima- have occurred. The interoperability issues that have been
tions, as well as the maximum active set approximations, raised ought to be handled by the infrastructure. Further-
are of interest to researchers. The implementation, on the more, it is vital to place equal focus on the simultaneous
other hand, is derived from theoretical derivations and is settlement of security issues in addition to these concerns.
thus derived, unlike theoretical derivations. According to Instead of focusing on anything else, the context-aware
the authors in Kempe et al. (2015), cloud-based computing

123
Imposition recognition in vehicular networks using hybrid offloading techniques...

judgements produced by rapidly developing networks are frequent topological changes because of the high mobility
customised to the proper fixed environment. of nodes in the network, which causes topological changes
to occur on a regular basis. Develop a creative method that
will efficiently transport the messages to their ultimate
3 Research methodology destination while adhering to the time limitations and other
constraints that have been defined. All the concerns have
In this part, we present a high-level overview of the been considered, and a unique routing mechanism has been
functional condition that the proposed approach is intended proposed that is the most efficient at routing packets from
to address. Figure 1 is an illustration of an example that any source to their intended destinations. When employed
may be used to offer a more straightforward explanation. in VANETs, the protocol is effective in both sparse and
The suggested architecture incorporates both a cellular densely populated environments. The major objective in
network interface and an IEEE 802.11p network interface, this scenario is to get a high PDR in the shortest period
resulting in each vehicle having two network interfaces. while incurring the least amount of administrative over-
Vehicles can communicate with the SDNi-MEC server and head possible. The difficulties associated with broadcast
transfer context over BSs that are equipped with the storm and grey zone, which were prevalent in most of the
wireless OpenFlow protocol (Lee et al. 2013). This context previous solutions, are no longer present with this protocol.
includes the vehicle’s position, which can be determined The proposed QADD system also recovers quickly in the
using GPS, speed, and direction, as well as the IDs of case of a connection breakdown, eliminating the issue of
surrounding cars that can be felt, that is, those cars that are fault tolerance that has previously been encountered. When
nearby and whose beacons can be picked up. The accom- using the mobility model, the highways are bidirectional,
panying execution strategy, which is based on the func- with two lanes on each side of the road in each direction. A
tional scenario shown in Fig. 1, is as follows: if at all total of N automobiles is present on the network, all of
possible, let the peer automobiles x and y connect with one which are travelling utilising the Freeway mobility para-
another over the cellular network before initiating com- digm. Listed below are the limits that were considered
munication. Vehicles x and y, as well as those who wish to throughout the design process. A binomial distribution is
do so, must report/transmit their contexts to the SDNi- used to disperse automobiles at junctions, with the proba-
MEC server on a regular basis to benefit from the V2V bility of making a left turn or a right turn at the same
VANET offloading feature. moment being equal (p). Traveling straight ahead increases
the likelihood of travelling in that direction by one to two
3.1 Data collection model points per second (1-2p). Veh x is made up of two parts: a
velocity component (vx) and an angle component (_x).
Although various cutting-edge routing protocols are Each of these parts is responsible for moving the vehicle.
available for VANETs, these solutions are inadequate for Before continuing further, we must first establish the fol-
delivering offloading to the large number of applications lowing lemmas to be true. Theory 1: The least amount of
that operate over these networks. VAENTs experience communication time is achieved between vehicles travel-
ling in opposite directions. Theory 2: As an example, Veh x
and Veh y are travelling in the communication range at an
angle of _x and _y with respect to one another and at a
relative velocity of as each.
  
v cosðhx Þ  vy cos hy i^
vrel ¼  x   : ð1Þ
vx sinðhx Þ  vy sin hy j^
The exact value of relative velocity is given as
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
jvrel j ¼ v2x þ v2y  2vx vy cos hx  hy : ð2Þ

Rx and Ry are the ranges of vehicles x and y, respec-

tively, in this example. Rx ? Ry is the maximum covered
area in which two vehicles may communicate at the same
time. The maximum amount of time t max during which
Veh h x and Veh may exchange messages is defined as

Fig. 1 Collection of data from vehicular network

123
 P. Bhanupriya, B. Ramachandran

Rx þ Ry based on connectivity time and offloading capacity of the

tmax ¼ : ð3Þ
jvrel j network. The effect of number of vehicles, RSUs capacity
and maximum speed has an impact on amount of offloaded
When the vehicles are travelling in the opposite direc- data. Regarding RSUs capacity, insufficient capacity leads
tion, the difference between their angles, i.e., _x- y = and to limited offloading of medium and low priority data,
cos(y) = - 1, causes Eq. (3) to become true, and the resulting in lower total data offloaded and higher maximum
vehicles are moving in the same direction. offloading delay. For effective complete offloading of data,
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
jvrel j ¼ v2x þ v2y þ 2vx vy : ð4Þ compression techniques are considered.

When using Eq. (4) to get the maximum transmission 3.2.1 Data offloading using hybrid particle swarm
time t max, the denominator obtained from Eq. (2) is the optimization with gray wolf optimization
maximum, resulting in t max being the lowest. As a result,
cars travelling in the other direction have the shortest A solution that is the best match in the GWO technique is
amount of communication time. referred to as alpha, whilst the second-best and third-best
solutions are referred to as beta and delta, respectively, in
Lemma 2 Vehicles travelling in the same direction have the algorithm. Omega solutions are the solutions that are
the most amount of communication time available. left over after all the other solutions have been exhausted.
Proof When Veh x and Veh y travel in the same direction, Using the algorithm, grey wolf hunting behaviour may be
the angle between them, i.e., _x- y = 0 and cos (0) = 1, is divided into three steps: surrounding prey, hunting, and
equal to the angle between them. Equation (5) calculates attacking prey (and looking for prey), and seeking and
the value of relative velocity as follows: searching for prey (and surrounding prey). Figure 2 of this
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi page depicts the overall processing unit in its entirety.
jvrel j ¼ v2x þ v2y  2vx vy : ð5Þ A basic component of the PSO algorithm was initially
proposed by Kennedy and Eberhart in Kempe et al. (2015),
and it was largely motivated by models of animal social
This is the bare minimum when computing t ‘‘max,’’ behaviour, such as bird flocking and fish schooling. It was
since the denominator is the smallest, the t max is the Kennedy and Eberhart that came up with the PSO algo-
largest. There are four channels available, each with a rithm, which was designed to imitate the social behaviour
different data rate ranging from 1 to 11 Mbps. Each vehicle of birds and fish. While searching for food, the birds may
is equipped with the infrastructure required to make use of disperse or concentrate to find a spot where they will have a
a variety of interfaces from these standards. The interfer- better chance of locating the food. The fact that one bird
ence ratio is the ratio of the number of links that are being can smell food extremely well while the others are moving
used to the number of connections that are available. With from one location to another in search of food indicates that
an increase in the interference ratio, throughput improves the bird is aware of where the food can be found and that it
and context switching latency reduces, which is due to has received the appropriate food resource communication
greater usage of the network because of increased network from the environment while the others are moving from
utilization. one location to another in search of food Because they are
broadcasting a message, particularly a valuable one, at any
3.2 Selection of offloading data parameters moment while moving from one site to another in search of
food, the birds will ultimately swarm to the location where
The data from the vehicle could be categorized into three food can be found as a consequence of the message they
types: immediate (accident and health), intermediate are transmitting. This is called a swarming effect.
(traffic condition, periodic health monitoring) and ordinary It is possible to calculate global optimization functions/
(delay-tolerant data). The overloading of the RSU is con- problems using this approach, which is taken from animal
trolled depending on the priority of data. Effective data behaviour, and each member of the swarm/crowd is
offloading will takes place only knowing the size of the referred to as a particle. The PSO technique, which uses
data being generated, minimizing latency for real-time two mathematical equations to identify their respective
applications, optimizing available bandwidth, ensuring locations, updates the location of each partner in the crowd
data reliability and security, scalability to accommodate in the global search space by using two mathematical
varying data volumes, adaptability to different network equations to calculate their relative positions. The mathe-
connectivity scenarios, energy efficiency to preserve matical equations presented in this section include the
vehicle battery life. The offloading requirement is decided following:

123
Imposition recognition in vehicular networks using hybrid offloading techniques...

Fig. 2 Overall processing unit

of proposed work

   
vkþ1 ¼ vki þ c1 r x pki  xki þ c2 r y gbest  xki answers to the challenges at hand. To develop the strength
i : ð6Þ
xkþ1 ¼ xki þ vkþ1 of both versions, we increase the ability of exploitation in
i i
Particle Swarm Optimization while concurrently boosting
We combine the particle swarm optimization algorithm the ability of exploration in grey wolf optimization.
with the grey wolf optimizer algorithm to generate a hybrid With the help of the mathematical equations that have
approach, which we call a low-level coevolutionary mixed been presented in HPSOGWO, the position of the first
hybrid method. Because we are combining the capabilities three agents in the search space is constantly being updated
of both kinds into a single object, the hybrid is considered (5). Instead of applying typical mathematical calculations,
low level. The fact that we do not use both forms in a we utilise the inertia constant to govern the exploration and
sequential manner implies that it is coevolutionary in nat- exploitation of the grey wolf in the search space as an
ure. In other words, they are both running at the same time. alternative to traditional mathematical equations.
As a result, it is classed as mixed since it has two distinct
versions that are both involved in the production of final

123
 P. Bhanupriya, B. Ramachandran

!  
d a ¼ !c1!
xaw!
x The hybrid optimization technique is used to identify
!   which task is better, and redundant data is removed
d b ¼ !c2!
xbw!
x : ð7Þ
!   because of the determination. It is conceivable that the
d d ¼ !c3!
xdw!
x optimized results achieved by the suggested approach
would give BS/AP with information that will enable it to
The following velocity and updated equation are pro-
allocate resources, including as power and sub-carriers, to
vided to integrate the PSO and GWO variants:
       task offloading in a more efficient manner. As a result, the
vkþ1
i ¼ w  vki þ c1 r 1 x1  xki þ c2 r 2 x2  xki þ c3 r3 x3  xki BS/AP will reply to the service requester (end user), who
xkþ1 ¼ xki þ vkþ1 :
i i will then make a choice on which offloading mechanism to
ð8Þ utilize based on the results of the investigation. The fol-
The Pseudo code of the proposed work is shown below. lowing is a more detailed description of the situation:

123
Imposition recognition in vehicular networks using hybrid offloading techniques...

which tasks should be computed locally, and which duties 3.3 Classification of offload parameters
should be offloaded to more resource intensive MEC ser-
vers is determined by the job that is assigned to the mobile This section details about the proposed model training to
device (MD) in question. Following that, the MDs contact identify the improper conditions based on speed, vehicle
the BS with which they are currently associated to get the alignment, discomfort noises. The proposed model consists
resources they need (home BS). The controller server, of the boosted feedforward layers that can achieve the
which is responsible for task scheduling and oversees better performances.
maintaining the resource allocation table (RAT), allocates
the most relevant MEC server within the cooperative area 3.3.1 Radial bias neural network (RBNN)
depending on the type of the query received at the base
station. Once the MDs have chosen which MEC server is The study article uses the ada-boost boosted, which oper-
the best appropriate for the job, they transfer the work to ates on the theory of extreme learning machines, to
that MEC server for completion. improve the performance of feed forward layers. The
Calculating the cost of a potential solution based on the classification accuracy of the boosting technique is
MDs’ existing positions is a challenging component of the improved by using the XG-boosting algorithm. By signif-
suggested Algorithm1. This comes on top of the chal- icantly revising the weights until the frameworks attain the
lenging calculations needed to update each particle’s highest level of classification/prediction accuracy without
velocity depending on Eq. (20) and the location of MDs sacrificing memory or time, the XG boost approach is
right now (21). The two processes that Algorithm1 opti- typically used to upgrade weak classifiers. The pseudo code
mizes for energy use and resource allocation are covered in for the boosting algorithm which is used in proposed net-
more detail later in this document. Before doing anything work is explained next.

further, the function should initialize the device’s locations, 3.3.2 Feedforwarded layers
which are represented as a dimensional vector of size k. In
this instance, t represents the initialization procedure’s Based on learning models Feedforward layers is designed.
maximum allowable temporal complexity (time complex- The input features maps of the ELM are denoted using
ity maximum). To find the optimal solution that fulfils the Eq. (9)
optimization objective in question, the search agents must X ¼ F ðF; PÞ; ð9Þ
first explore the solution space until they find it. As a result,
the complexity of computing the cost of candidate solu-
tions for each iteration is t max’, where k is the dimension
of the problem and x i(k ? 1) is the size of the population
of candidates for consideration.

123
 P. Bhanupriya, B. Ramachandran

where X is the fused spatio-temporal features obtained Admission control: An RSU has the authority to stop
from the CNN and GRU layers, F is the CNN spatio fea- servicing a vehicle that is currently offloading its data to
tures and P is the GRU temporal features. provide service to another vehicle. When a Vehicle V
The output ELM function is denoted by Eq. (10) having high-priority data wants to offload its data to RSU
 1 K at the same time when the current load C k at RSU K is
T 1 T equal to the maximum tolerable load Cmax K , the RSU will
Y ðnÞ ¼ X ðnÞb ¼ X ðnÞX XX O: ð10Þ
C stop the current offloading and will do the high-priority
data offloading by high-priority data.
The overall training of ELM is given by Eq. (11)
!
XN
3.4 Proposed model
S¼a ðY ðnÞ; BðnÞ; W ðnÞÞ ; ð11Þ
n¼1
Figure 3 represents the proposed framework architecture.
where X (n) is key feature, b tem[poral, C is constant, The proposed framework is a hybrid learning framework
B bias and W = weights. which significantly integrates the ELM frameworks with
the Boosted learning algorithm to achieve better perfor-
3.3.3 Quality of service functions (QoS) mance. The boosted learning model makes the proposed
classifier is so strong to prove the better performance.
Three quality of service (QoS) functions namely traffic Hence, proposed prediction framework imparts boosting to
classification, overload control and admission control are train the weights of ELM to boost its accuracy with training
used in the proposed system for the management of data data set. The input data that are used for the stored in the
offloading. cloud are used as the inputs to the proposed model. The
Traffic control: A vehicle first offloads high-priority mathematical formula for determining the ELM’s classifi-
data, then the medium priority data and finally the low cation outputs is given by Eq. (4)
priority data. If it has multiple data with the same priority, !
XN
it uses a greedy approach and first offloads the data having S¼a ðY ðnÞ; BðnÞ; W ðnÞÞ : ð12Þ
the largest size. n¼1
Overload control: Use of overload control is to maintain
In the next step, calculated S is compared with threshold
the load of RSUs and keep their resources available for
which is given as Eq. (5)
future vehicles having high-priority data. For this purpose,
an offload criterion by considering threshold values of the SðtÞ ¼¼ S: ð13Þ
maximum allowed load at RSU K for medium and low If S is not equal to S (t), then boosting algorithm tunes
priority data is given for each vehicle. RSU K will allows a the bias weights for every iteration and it is given mathe-
Vehicle V having medium priority data to offload its data if matically by Eq. (6)
the current load is below the threshold value.

Fig. 3 Network architecture of

the proposed boosted model

123
Imposition recognition in vehicular networks using hybrid offloading techniques...

!
X
N 4.4 Obstacle shadowing and vehicle shadowing
Sðk þ 1Þ ¼ a ðY ðn þ kÞ; Bðn þ kÞ; W ðn þ kÞÞ ð14Þ
n¼1 The vehicle data signal is also affected by the signals
Where k ¼ 1; 2; 3; . . .. . .N  1 until S ¼¼ SðtÞ passing through buildings and vehicle, which is represented
by polygons. Considering the path loss, two ray path loss
obstacle shadowing and vehicle shadowing the received
power is given by,
4 Mobility parameters
Power received ¼ Treansmitted powerð20mwÞ
4.1 Attenuation and path loss þ sending antenna gain
þ receiving antenna gain  path loss:
Data transfer in a vehicle communication happens in ð17Þ
atmosphere and as the data travels in space, it becomes
weak with respect to increase in distance and this factor is
4.5 SNR, BER and the decider
called attenuation or path loss.
Power transmittedðPt Þ The transfer of data packets happens through all the nodes.
Pathloss ¼
Power receivedðPr Þ The decider identifies the successful reception of data using
ð4  pi  5:9  dÞ2 the class [Link] signal to noise ratio (SNR)
¼ 10log : ð15Þ
c2 occurs when two nodes initiate the transfer of data to one
The transmitted and received power and measured in node at the same time.
terms of watts. The vehicular communication uses Wifi Signal to interference ratio (SINR)
standard of 802.11p and Wireless Access in Vehicular Recieved powerðPÞ
SINR ¼ P : ð18Þ
Environments (WAVE) uses the carrier frequency of Noise þ Interfering packets power
5.9 GHz. The d is the distance travelled by the signal in
Lower SINR represents means the noise is more than the
terms of meters and speed of light is given as C. The path
signal while the higher SINR Bit error ratio (BER) is
loss is measured in terms of decibels.
fraction of change in bits count to the number of bits
transferred for a period. Bit error rate is inversely propor-
4.2 Received power
tional to SINR and power of the signal. Therefore, the
higher the data rate, the more power is needed to ensure
Other than path loss and attenuation the other factors
that each bit is readable.
affecting the data transfer are radiated power, possible
gains and losses. The complete received power is given by
sum of these values.
5 Results and analysis
Power received ¼ Transmitted powerð20mwÞ
þ sending antenna gain In this section, performance of the proposed model is
þ receiving antenna gain  path loss evaluated based on the obtained pre-processed data
ð16Þ (Fig. 4). For the further evaluation of the result, different
activation functions are applied on the model. Table 1
The transmission power is set to 20 mw. The antenna
presents the different performance metrics used for the
shape, vehicle speed, position of vehicle and orientation
detection of emergency and evaluation of the proposed
determines the antenna gain.
model with incorporation for additional solution from [31].
Figure 5 shows the number of offloading users on dif-
4.3 Two-way path loss
ferent datasets. Table 2 presents the mathematical expres-
sions of performance metrics.
The constructive interference and destructive interference
Figure 6 shows validation performance for delay com-
is caused due to the data signal that is reflected by the
parison and Fig. 7 shows the effect of vehicles on average
ground during transfer resulting in two-way path loss. As
delay. From the figures, the error between the training and
the distance increases the two-way path loss also increases.
validation performance is found to be less as equal to 0.001.

123
 P. Bhanupriya, B. Ramachandran

Fig. 4 Surface diagram of

optimized algorithm

Table 1 Training parameters used for the proposed model Table 2 Mathematical expressions for performance metric’s
calculation
S. no. Detailed parameters Specifications
S. no. Performance metrics Mathematical expression
01 No of Epochs 100
01 Accuracy T:PosiþT:Neg
02 Batch size 20 T:PosiþT:NegþF:PosiþF:Neg
03 Learning rate 0.0001 02 Re-call T:Posi
 100
T:PosiþF:Neg
04 Training data 70 03 Specificity T:Neg
T:NegþF:Posi
05 Testing data 30
04 Precision T:Neg
T:PosiþF:Posi
05 F1-score PrecisonRecall
PrecisionþRecall

T:Posi is true positive values, T. Neg true negative values, F:Posi is

false positive, F. Neg is false negative values

found that the AUC for detecting the three energies

maintains the AUC (area under curve of 0.975) and AUC
of 0.96 in detecting the abnormal conditions.
To prove the stability of the model, different activation
functions are adopted for model training and performance
metrics are calculated. Figure 10, Fig. 11, Fig. 12 and Fig. 13
present the performance metrics of the proposed model with
the different activation functions. To prove the excellence of
the proposed algorithm, performance of the model is com-
pared with the other state-of-art learning models such as
‘‘Support vector machines (SVM) [32], K-nearest neighbor-
hood (KNN) [33], artificial neural network (ANN) [34], ran-
dom forest (RF) [35], Naı̈ve Bayes (NB) [36], extreme
learning machines (ELM) [37] and boosting algorithm [38]’’.
Fig. 5 Number of offloading users vs number of users
Tables 3, 4, 5, 6 present the relative analysis of the
different algorithms in various condition of road. The
Confusion matrix is a clean representation of the models’
hybrid model has better performance in contrast with the
performance in terms of classification and prediction.
other methodologies.
Figure 8, ROC performance and Fig. 9 represents
energy consumption of vehicles during data offloading. It is

123
Imposition recognition in vehicular networks using hybrid offloading techniques...

Fig. 6 Validation performance in detecting normal condition

Fig. 9 Energy consumption reduction on vehicle data offloading

Fig. 7 Delay comparison for increase in vehicle
boosted ELM and a novel dataset collection. The suggested
6 Conclusion algorithms and current algorithms are subjected to perfor-
mance evaluation to determine the efficiency with which
After applying boosted ELM and a new dataset collection, the offered approaches perform. Because the suggested
the proposed framework’s goal is to forecast offload Hybrid PSO with GWO achieves the best accuracy of
parameters and eliminate mismatch parameters by utilising 95.26% for server selection, it can be concluded that the

Fig. 8 ROC performance for the

proposed model optimization

123
 P. Bhanupriya, B. Ramachandran

Fig. 10 Performance metrics of

the proposed model using the
Tanh activation function

Fig. 11 Performance metrics

using the relu activation
function

Fig. 12 Performance metrics

using the sigmoidal activation
function

Fig. 13 Performance metrics of

the proposed model using the
RBF activation function

123
Imposition recognition in vehicular networks using hybrid offloading techniques...

Table 3 Relative analysis of the

S. no. Algorithms Performance metrics
different algorithms in detecting
the normal energies Accuracy Precision Re-call Specificity F1-score

01 KNN 82% 81.2% 81.0% 80.2% 80.3%

02 ANN 72.4% 72.33% 72.48% 72.35% 72.47%
03 NB 68.2% 67.9% 68.2% 67.3% 67.9%
04 RF 80.0% 73.9% 73.9% 74.9% 75.2%
05 BOOSTING ALGORITHM 78.4% 77.2%% 79.4% 78.3% 78.2%
06 SVM 88.2% 87.92% 86.93% 87.5% 88.2%
07 ELM 89.1% 88.56% 88.4% 88/7% 88.2%
08 PROPOSED MODEL 95.1% 95.2% 95.0% 95.1% 95.12%

Table 4 Relative analysis of the

S. no. Algorithms Performance metrics
various algorithms in detecting
the pitta disorder energies Accuracy Precision Re-call Specificity F1-score

01 KNN 82% 81.2% 81.0% 80.2% 80.3%

02 ANN 72.4% 72.33% 72.48% 72.35% 72.47%
03 NB 68.2% 67.9% 68.2% 67.3% 67.9%
04 RF 80.0% 73.9% 73.9% 74.9% 75.2%
05 BOOSTING ALGORITHM 78.4% 77.2%% 79.4% 78.3% 78.2%
06 SVM 88.2% 87.92% 86.93% 87.5% 88.2%
07 ELM 89.1% 88.56% 88.4% 88/7% 88.2%
08 PROPOSED MODEL 95.1% 95.2% 95.0% 95.1% 95.12%

Table 5 Relative analysis of the

S. no. Algorithms Performance metrics
different algorithms in detecting
the emergency conditions Accuracy Precision Re-call Specificity F1-score

01 KNN 82% 81.2% 81.0% 80.2% 80.3%

02 ANN 72.4% 72.33% 72.48% 72.35% 72.47%
03 NB 68.2% 67.9% 68.2% 67.3% 67.9%
04 RF 80.0% 73.9% 73.9% 74.9% 75.2%
05 BOOSTING ALGORITHM 78.4% 77.2%% 79.4% 78.3% 78.2%
06 SVM 88.2% 87.92% 86.93% 87.5% 88.2%
07 ELM 89.1% 88.56% 88.4% 88/7% 88.2%
08 PROPOSED MODEL 95.1% 95.2% 95.0% 95.1% 95.12%

Table 6 Relative analysis of the

S. no. Algorithms Performance metrics
different algorithms in detecting
the Emergency Conditions Accuracy Precision Re-call Specificity F1-score

01 KNN 82% 81.2% 81.0% 80.2% 80.3%

02 ANN 72.4% 72.33% 72.48% 72.35% 72.47%
03 NB 68.2% 67.9% 68.2% 67.3% 67.9%
04 RF 80.0% 73.9% 73.9% 74.9% 75.2%
05 BOOSTING ALGORITHM 78.4% 77.2%% 79.4% 78.3% 78.2%
06 SVM 88.2% 87.92% 86.93% 87.5% 88.2%
07 ELM 89.1% 88.56% 88.4% 88/7% 88.2%
08 PROPOSED MODEL 95.1% 95.2% 95.0% 95.1% 95.12%

123
 P. Bhanupriya, B. Ramachandran

vehicle data offloading has been completed with the most Eagle N, Pentland A, Lazer D (2007) Inferring social network
appropriate server by using this method. After that, tech- structure using mobile phone data. Proc Nat Acad Sci
106(36):15274–15278
niques like clustering and channel selection are used to Gupta V, Rohil MK (2012) Mobile data offloading: benefits issues
ensure that data transfer is carried out in an energy-efficient and technological solutions. In: International Conference on
way. Furthermore, when compared to existing scheduling Computer Science Engineering & Applications (ICCSEA),
algorithms, the energy consumption cost produced by the pp 73–80
Gao G, Xiao M, Wu J, Han K, Huang L (2016) Deadline-sensitive
suggested scheduling method is quite low. As a result, the mobile data offloading via opportunistic communications. In:
tasks that have been submitted are completed in the Proc. 13th Annu. IEEE Int. Conf. Sens. Commun. Netw.
shortest amount of time feasible. Furthermore, new tech- (SECON), pp 1–9
nologies for detecting insider attacks on vehicle networks Gauni S, Bhanupriya P, Kalimuthu K et al (2023) Correlation and
contrast of multi-user edge computation with single-user edge
will be used in the future, and the preventative mechanism computation for data offload on terrain electric vehicular
will be equipped with a high level of safety. applications. Multimed Tools Appl 82:26563–26575. https://
[Link]/10.1007/s11042-023-14848-6B
Ghorbani M, Rabiee HR, Khodadadi A (2016) Bayesian overlapping
Author contributions The manuscript was written through contribu- community detection in dynamic networks. arXiv preprint
tions of all authors. All authors have given approval to the final [Link]
version of the manuscript. Han B, Hui P, Kumar V, Marathe MV, Pei G, Srinivasan A (2010)
Cellular traffic offloading through opportunistic communica-
Funding No funds, grants, or other support was received. tions: a case study. In: Proc. of the 5th ACM workshop on
Challenged networks (ACM CHANTS 2010), pp 31–38
Availability of data and materials Not applicable. Han B et al (2012) Mobile data offloading through opportunistic
communications and social participation. IEEE Trans Mob
Code availability No. Comput 11(5):821–834
Hui P, Chaintreau A, Scott J, Gass R, Crowcroft J, Diot C (2005)
Pocket switched networks and human mobility in conf environ-
ments. In: Proc. ACM Special Interest Group Data Comm.
Declarations Workshop (SIGCOMM’05)
Jiang J, Zhang S, Li B (2016) Maximized cellular traffic offloading
Conflict of interest The authors have no financial or proprietary via device-to-device content sharing. IEEE J Sel Areas Commun
interests in any material discussed in this article. The authors declare 34(1):82–91
that they have no conflict of interest. Kempe D, Kleinberg JM, Tardos E (2015) Maximizing the spread of
influence through a social network. Theory Comput
Ethical approval Not applicable. 11(4):105–147
Lee K, Lee J, Yi Y, Rhee I, Chong S (2013) Mobile data offloading:
Informed consent Not applicable. how much can WiFideliver? IEEE/ACM Trans Netw
21(2):536–550
Liu W, Gong W, Du W, Zou C (2017) Computation offloading
References strategy for multi user mobile data streaming applications. In:
19th International Conference on Advanced Communication
2021 Global Networking Trends Report. [Online]. Available: https:// Technology (ICACT), pp 111–120
[Link]/c/en/us/solutions/executiveperspectives/annual- Loukas G, Yoon Y, Sakellari G, Vuong T, Heartfield R (2017)
internet-report/[Link]. Accessed 6 Dec 2020 Computation offloading of a vehicle’s continuous intrusion
Ackerman E, Ben-Zwi O, Wolfovitz G (2010) Combinatorial model detection workload for energy efficiency and performance.
and bounds for target set selection. Theoret Comput Sci Simul Model Pract Theory 73:83–94
411(44–46):4017–4022 Orimolade J, Ventura N (2015) Intelligent access network selection
Andreev S, Pyattaev A, Johnsson K, Galinina O, Koucheryavy Y for data offloading in heterogeneous networks. In: AFRICON
(2014) Cellular traffic offloading onto network-assisted device- 2015, pp 1–5, 107
to-device connections. IEEE Commun Mag 52(4):20–31 Shahryari O-K, Pedram H, Khajehvand V, Takht-Fooladi MD (2021)
Bhanupriya P, Gauni S, Kalimuthu K et al (2021) A modified hybrid Energy and task completion time trade-off for task offloading in
blockchain framework for secured data transaction. J Phys: Conf fog-enabled IoT networks. Pervasive Mob Comput 74:101395
Ser 1964:042040 Spatharakis D, Dimolitsas I, Dechouniotis D, Papathanail G, Fotoglou
Bhanupriya P, Gauni S, Kalimuthu K et al (2022) Knowledge I, Papadimitriou P, Papavassiliou S (2020) A scalable edge
discovery of edge computation for offload vehicular applications computing architecture enabling smart offloading for location
in IoT. Wirel Pers Commun 126:2347–2359. [Link] based services. Pervasive Mob Comput 67:101217
1007/s11277-021-09191-8 Srinivasan V, Motani M, Ooi WT (2006) Analysis and implications of
Cisco Annual Internet Report (2018–2023) White Paper. Accessed: student contact patterns derived from campus schedules. In:
Dec. 6, 2020. [Online]. Available: [Link] Proc. The 12th ACM International Conference on Mobile
us/solutions/collateral/executiveperspectives/annual-internet- Computing and Networking (MOBICOM’06), pp 86–97
report/[Link] Valerio L, Bruno R, Passarella A (2014) Adaptive data offloading in
Cisco Visual Networking Index: Global Mobile Data Traffic Forecast opportunistic networks through an actor-critic learning method.
Update, 2017–2022. In: Cisco white paper, 2019. In: Proceedings of the 9th ACM MobiCom workshop on
Challenged networks (ACM CHANTS), 2014, pp 31–36

123
Imposition recognition in vehicular networks using hybrid offloading techniques...

Valerio L, Bruno R, Passarella A (2015) Cellular traffic offloading via Springer Nature or its licensor (e.g. a society or other partner) holds
opportunistic networking with reinforcement learning. Comput exclusive rights to this article under a publishing agreement with the
Commun 71:129–141 author(s) or other rightsholder(s); author self-archiving of the
accepted manuscript version of this article is solely governed by the
Publisher’s Note Springer Nature remains neutral with regard to terms of such publishing agreement and applicable law.
jurisdictional claims in published maps and institutional affiliations.

123