VANET: A Machine Learning Approach

*Omkar Pattnaik, Sasmita Pani and Binod Kumar Pattanayak*

#### **Abstract**

Emerging automotive networks should make the automotive work more secure, greener, and more efficient, and clear the way for autonomous driving to the arrival of the 5G cell. Various new irritants arise due to high sequence factors in vehicle conditions and that is why reconsideration of remote design approaches is also considered. The smart vehicle of the future, which are the backbone of high-performance multi-purpose networks, are gradually being given more advanced sensors and continue to produce information volumes. Machine learning (ML), as an integral part of artificial intelligence (AI), has recently been used in wireless networks to provide continuous information to deal with culturally challenging issues. In this chapter, we examine the recent progress in the use of machine learning in automotive networks and try to pay close attention to this emerging domain. We began to differentiate the obvious characteristics of multi-functional vehicle networks and promote the use of a learning machine to deal with subsequent difficulties.

**Keywords:** VANET, machine learning (ML), artificial intelligence (AI), autonomous driving, multi functional

#### **1. Introduction**

Despite the fact that VANET is derived from the concept of MANET, there are some subtle contrasts between them. A portable remote device is used to communicate between nodes in a MANET, but in the case of a VANET, running vehicles talk to other vehicles. VANETs are made of vehicles and highway guides. It is a critical method of a competent moving framework. The development of nodes in VANET is traditional as vehicles move along established roads with high momentum limited to the bottom of situations such as small or curved street, congestion/congestion and so on.

In designing this environment, we assume that each vehicle must be associated with an on-board unit (OBU) and a road-side unit (RSU) installed along the roads. RSUs and OBUs are required to communicate and exchange information with each other through a protocol called Dedicated Short-Range Communications (DSRC) [1]. Three types of communication are possible in VANET such as Vehicle to Vehicle (V2V), Vehicle to Infrastructure (V2I) and Infrastructure to Infrastructure (I2I). But V2V method uses DSRC protocol. It is maintained purely as an ad hoc network. The main goal of DSRC is to ensure low communication delay and high data transmission in the network [2]. The Federal Communications Commission (FCC) subsequently updated the DSRC protocol to WAVE (Wireless Access for Vehicular Environment).

**Figure 1.** *Structure of VANET.*

The pattern of VANET Structure is given in **Figure 1**.

Remote network which can support large portability broadband get to have gotten increasingly more consideration from both industry and the scholarly community as of late [3, 4]. Specifically, the idea of associated vehicles or vehicular systems, as appeared in **Figure 2**, has increased considerable energy to bring another level of availability to vehicles and, long side novel onboard figuring and detecting advancements, fill in as a key empowering influence of Intelligent Transportation System (ITS) and developed urban areas [5]. This new age of networks will at last have a significant effect on the society, making ordinary traveling more secure, greener, and progressively productive and comfortable. Alongside recent advances in a wide scope of artificial intelligence (AI) advances, it is helping clear the way to autonomous driving in the appearance of the fifth generation cell frameworks (5G).

Meanwhile, scheduled inventive vehicles are progressively furnished with a broad assortment of sensors, for example, "engine control units, radar, light detection and ranging (LIDAR), and cameras" to enable the vehicle to see the general climate just as observing its individual activity quality in current world. Along with superior concluding and capacity gadgets onboard, these sensible innovations are switching vehicles from a basic remodel efficiency to a capable figuring and connections centre with rational handling potentiality. They continue gathering, producing, putting away, and conveying large volumes of information, subject to additionally processing

**Figure 2.** *Structure of VANET with LIDAR.*

#### *VANET: A Machine Learning Approach DOI: http://dx.doi.org/10.5772/intechopen.109349*

and usually called as mobile big data [6]. Such information give rich context data in regard to the vehicle energy, (for example, speed, acceleration and direction), street conditions, traffic flow, wireless environments, and so on that can be motivated to improve network performance through versatile data driven dynamic. Generally, traditional interchanges techniques are not intended to deal with and exploit such data.

The applications produced for VANETs can be grouped into 3 principle classes: "(1) Safety applications (e.g., road risk control warning and emergency electronic break light), (2) comfort applications (e.g., parking accessibility notice and blocked road notice), and (3) commercial applications (e.g., service announcements and content map database download) [7]." These programs produce two types of messages for VANET exchanges, which include secure information and non-secure information. Safety information, including reference points and urgency information, travel on the control path. Non-secure information, including information generated by convenience or business applications, are routed to the favorable path [8, 9].

In city conditions, the junctions are imperative locations. So these are the often probably positions for traffic accident. In "Canada, nearly 800 avenue customers killed and 7250 critically injured at intersection clashes in 2005 [10]." The latest statistical information of avenue clashes mentioned via "Canada Road Safety Vision 2010 indicates that approximately 47% of all human beings killed and 57% of all injured [3]." Moreover, primarily located on the collective facts of avenue clashes in "Brampton roadways, between 2003 and 2007, 71% of pedestrian collisions came about at intersections [3]." To furnish safety and greater reliable surroundings about road customers on the junction of city roads, the use of VANETs appear to be crucial.

An eminent level of "Quality of Service (QoS)" is essential at junctions to prevent any verbal exchange conflicts that might also appear due to fleshy conversation bundle. Development "of QoS in VANETs" is a demanding assignment due to may specific VANET features such as topology settings and common damaged paths [11].

#### **2. Related work**

Machine learning methods can gain from a dataset via naturally breaking down the dataset to recognize rules. We can utilize these principles to anticipate results from new information. When all is said in done, machine learning frameworks need a pretraining procedure with the end goal that they can generalize input information. As indicated by various training information, machine learning procedures can be arranged as either unsupervised clustering or supervised classification. Unsupervised machine learning frameworks analyze the similarity between information to choose whether such information can be gathered. Supervised machine learning frameworks map input information to the ideal outputs, characterizing input information as per their qualities.

Currently machine learning frameworks have been utilized to improve the presentation of wireless networks. When packets are lost in the traditional TCPaccommodating rate control protocol, it is viewed as network congestion. The protocol will decrease the bundle transmission rate and lower channel usage. In Ref. [12], the authors use machine figuring out how to decide the reason for packets misfortunes and take various measures to improve the system execution. The reason can be congestion, a path change, or connection errors. The cluster-based routing rules can assemble nodes into clusters and recognize cluster heads and general hubs through machine learning procedures [13]. In Ref. [14] communicate routing utilizes machine learning to foresee whether a packet should be rebroadcast. In Ref. [15], the authors

powerfully alter the reference point interval through machine learning, which diminishes the control overhead and keeps up the unwavering quality of transmissions.

With regard to vehicular correspondences, the authors explain information communicated over the system originates from various sensors and On-Board Units (OBUs). Mischief recognition, explicitly information driven trouble making identification, is a significant piece of the security framework. It centres on choosing the accuracy of information communicated and contrasting it and past data received. The greater part of the work directed in information driven either requires agreement [16] or the presence of a central authority [17]. Introduced an adaptable central misbehavior assessment framework for OBUs and roadside infrastructure units that identifies and dispenses with aggressors from the system dependent on information believability. The methodology included each node, recognizing an incident, to advance report to the central power to take a ultimate choice.

K-Nearest Neighbor (K-NN): K-NN evaluation is a common anomaly detection/ classification technique and broadly used due to its easy implementation and capability to naturally cope with multiclass cases [18].

In this case, authors have mentioned regular K-NN which analyses the conduct of statistics factors in accordance to its previously broadcasted role primarily based on "Euclidean distance." "A data point is labelled as an outlier if it is situated farther away from its earlier broadcasted points. In Third International Knowledge Discovery and Data Mining Tools Competition, KDD cup 1999 dataset was once used to build a IDS the use of classification algorithms. The 1-nearest neighbour classifier algorithm, 1-NN, performed similar to the winning entry and proven magnificent results in terms of detection" [19].

"Li et al. [20] investigated the effect of the Flooding Attack in wi-fi sensor networks and introduced an intrusion detection device based on K-NN algorithm. The K-NN algorithm" was correctly in a position to observe the ordinary nodes using the frequency of sent messages as a characteristic vector.

Support Vector Machine (SVM): In Ref. [6] the authors trying to focus on SVM is a discriminative classifier, characterized by an isolating hyper plane, that maps the information focuses on the plane. This direct machine has a place with the class of Kernel techniques with issue explicit portion work. The job of Kernel work in "SVMs is to prompt an element space by mapping the training information on a high dimensional space where it is straight distinct. SVM have shown achievement in the field of text mining, pattern recognition, object classification, and anomaly detection. Ref. [10] displayed classification" of various kinds of interruptions utilizing supervised learning procedures on the ISC2012 dataset, a well-known benchmark dataset for interruption location [2]. The work exhibited that "SVM, with a discovery precision of 99.1%, could be utilized effectively as an classification module of a network anomaly IDS".

"In Ref. [21], the author showed a cloud-based mobile distributed intrusion detection system for VANET that detects Denial-of-Service attack using One Class SVM (OCSVM) classifier with RBF kernel and score the peculiarities utilizing recursive K-implies clustering. The framework exactness was compared on the quantity of transmitted reference point by vehicles, the speed of intruder vehicle, and the separation between Road Side Units".

#### **3. Vehicular networks facing difficulties with high mobility factor**

High portability vehicular systems show unmistakable qualities, which have presented critical "provocation to wireless network" structure. In current segment, we recognize alike difficulties and afterward examine the capability of utilizing "machine learning" into the address domain.

#### **3.1 Stable mobility**

"Higher flexibility of vehicles leads to stable dynamics and affects machine design in multiple aspects of the connections network. One of the most important ways that high mobility sets itself apart is through its unique channel propagation component networks compared as low mobility counterparts. For example, vehicular channels exhibit rapid sensual variation and also suffer from inherent non-stationary of carrier statistics as on their different natural environment effective" [7, 22]. "This one is made worse by the fact that carrier statistics, which are typically used to increase evaluation accuracy, are non-stationary" [8].

"Meanwhile, due to then high Doppler spread caused by vehicle mobility, then multicarrier modulation scheme is maximum susceptible into inter carrier interference (ICI) in vehicular networks" [10, 11] and hence brings difficulty to signal identification. Constant movement of vehicles also causes usual modifies of then communications network arrangement, affecting channel allocation and routing protocol designs. For instance, in cluster-based vehicular networks [12], "moving vehicles may combine and leave the cluster repeatedly, making it hard to maintain long-lasting communications within then formed cluster and so warranting further study on cluster stability. Another sources of dynamics in higher flexibility networks comes from then unstable vehicle density, which modify dramatically reliant on then locations" and moment "(peak or off hours of the day)". "Flexible and robust resources management plan that makes an efficient use to available resources while adapting to then vehicle density changing are thus required".

#### **3.2 Independent and demanding QoS provisions**

In large portability vehicular systems, there may be chances of various sorts of links, that we comprehensively order "into V2I and V2V connections. The V2I" joins empower vehicles to convey "with the base station to help different traffic efficiency and data (infotainment) services". "They generally require visit access to the Web or remote servers for media streaming, map downloading, and social" communication, that include impressive measure of information convection and in this way are maximum transfer speed concentrated [5]. Then again, the V2V joins are for the most part considered for sharing safety basic data, as basic safety messages (BSM) in DSRC [23], among vehicles in close proximity in either an intermittent or occasion activated way. Such safety related messages are carefully defer touchy and require high unwavering quality.

Specifically, "reinforcement learning [24], one of the machine learning devices, can collaborate with the dynamic" condition and create palatable customs to satisfies differing QoS prerequisites about vehicular systems during adjusting to shifting mobile condition. For instance, "in resource allocation problems, the optimal strategies are first learned and afterward the vehicle agents" in like manner catch activities to modify controls and appropriate channels versatile to the changing situations portrayed by, e.g., link conditions, privately saw obstruction, and vehicle energy while conventional static numerical models are not so good at catching and tracking such powerful changes [25, 26].

#### **4. Introduction to machine learning**

Machine learning (ML) permits computers to discover concealed bits of knowledge through iteratively gaining from information, without being expressly customized. It has changed the universe of computer science by permitting learning with enormous datasets, which empowers machines to change, re-structure and advance algorithms without anyone else. Existing machine learning strategies can be separated into 3 classes, to be specific, supervised learning, unsupervised learning, and reinforcement learning. Other learning plans, for example, semi-administered learning, web-based learning, and transfer learning, can be seen as variations of these three essential sorts. When all is said in done, ML includes 2 phases, i.e., training and evaluating. "In the training stage, a model is found out dependent on the training" information "while in the testing stage, the trained model is applied" to create the expectation. In current segment, we quickly present the idea of" ML with the expectation which then researchers can value may be in taking care of generally testing issues (**Figure 3**).

#### **4.1 Supervised learning technique**

The popularity of practical ML algorithms utilize supervised learning with a labeled dataset, where each preparation test accompanies a label. A definitive objective "of supervised learning is to discover the mapping from the input data space to the label" with the goal that dependable expectation could be done when fresh knowledge is provided. Issues with supervised learning can also be divided into "classification and regression," in which the contrast the difference between the two projects is that the labels for classification are absolute while those for regression are numerical. Allocation procedures figure out how to anticipate a classification output for every approaching sample dependent on the training information. Some classic procedures in this category incorporate Bayesian classifiers [15], k-nearest neighbors (KNN) [27], decision trees [14], support vector machine (SVM) [16], and neural networks [28]. Rather than discrete outputs, regression procedures predict a consistent value relating to each sample, for example, estimate element inputs. Classic regression procedures incorporate logistic regression [24], support vector regression (SVR) [17], and Gaussian process for regression [29].

#### **Figure 3.**

*Classification of machine learning structure.*

#### *VANET: A Machine Learning Approach DOI: http://dx.doi.org/10.5772/intechopen.109349*

"If the training data only includes discrete values then it is a classification problem and the output of the trained model is a classification which is also discrete". "On the other hand if the training data containing continuous values, then it is the regression problem and the output of the trained model will be a prediction. Two widely used examples of supervised ML are Decision Trees and Random Forest".

The result of reverting procedures is a constant worth that may represent to forecast of the house value, stock trade, banking client exchanges, electric vehicle battery, level of traffic blockage at different crossing points, and jamming expectation (**Figure 4**).

In vehicular social network, regression can be utilized to foresee parameters, for example, network throughput. Two classic regression procedures are:1) Gaussian Process for Regression (GPR) and 2) Support Vector Regression (SVR).

In vehicular systems the classification procedures can be utilized for interruption or malfunction detection. Besides, classification procedures are additionally beneficial in the traffic secure applications, for example, Augmented Reality Head Up Display (AR-HUD), dynamic driver data frameworks, obstacle recognition, and anticipating complex traffic types.

#### **4.2 Unsupervised learning technique**

In the unsupervised ML, training depends on unlabelled information. "This plan trying to find an efficient portrayal of the unlabelled information. For instance, the highlights of data can be caught by various hidden factors that can be spoken to by the Bayesian learning methods". Clustering is a kind of unsupervised "discovering that gatherings tests with comparable highlights". The input information of each data point can be its supreme portrayal or a relatives similarity level into other information points (**Figure 5**).

In the wireless methods universally, the cluster organization for the hierarchical protocols is important with respect to energy management, where every node simply needs to speak with the cluster head before speaking with the individuals from different clusters. Some traditional clustering methods are k-means, spectrum clustering, and hierarchical clustering. Measurement reduction is another subclass of unsupervised ML conspire. The principle thought behind measurement reduction is to down-sample the information from a higher measurement to a lower measurement with no significant information loss [19]. Implementing machine learning for most applications require measurement reduction because of various reasons.

**Figure 4.** *Supervised learning structure.*

*Vehicular Networks – Principles, Enabling Technologies and Modern Applications*

#### **Figure 5.**

*Unsupervised learning structure.*

A delegate instance of unsupervised learning is clustering, to be specific, to assemble tests such that samples in a similar cluster have a larger number of similarities than the samples in various clusters. The highlights utilized for clustering could be either the outright portrayal of each sample or the relative similarities between samples [20].

Other than clustering, measurement reduction is another significant instance of unsupervised learning, where tests from a high dimensional space are anticipated into a lower one without losing an excess of data. In numerous situations, the raw information accompany high measurement, which is not possible due to a few reasons. One explanation is the purported revile of dimensionality [18], which depicts the hazardous marvel experienced when the measurement gets larger. For example, in enhancement, clustering, and classification, the model unpredictability and the quantity of required training samples develop significantly with the element measurement.

Reapproach information dimensionality is the first reason. In clustering, classification and advancement, the ordinary model unpredictability increasing significantly within the expansion of highlight instruments. The second reason is the obstacle in the learning procedure. In the greater part of the cases the highlights of the information samples are correlated in certain angles, yet on that the component value is changed by noise or impedance, at thats position the separate outcome of the "correlation will be depraved and the learning procedure will be influenced. Such type of measurement decreasing in the vehicular informal network is the development which prompts a vehicular cluster". Cluster head gathers and transform the data to the node B to diminish the correspondence costs. The curse of dimensionality can be decreased by the measurement decrement techniques. Measurement reduction strategies are gathered in two classes: "(1) linear projection methods such as: Principal Component analysis (PCA) and Singular Value Decomposition (SVD), and (2) nonlinear projection methods such as: manifold learning, local linear embedding (LLE) and is metering mapping" [21].

#### **4.3 Reinforcement learning technique**

Reinforcement learning in effectively gains from the activities of the taking in operator from the relating reward. It implies so as to expand the reward, inexplicit mapping the circumstances as per the activities by experimentation. The Markov Decision Process (MDP) is a case of reinforcement learning. Q function model-free learning process is an exemplary guide to take care of MDP advancement problem that does not require data about the learning condition.

*VANET: A Machine Learning Approach DOI: http://dx.doi.org/10.5772/intechopen.109349*

#### **Figure 6.**

*Reinforcement learning structure.*

In reinforcement learning issues, an operator learns the ideal practices through interfacing with nature in an experimentation way intending to augment compensations from nature. The environment is displayed as a Markov decision process (MDP), which acquaints activities and rewards with a Markov procedure. Both the state progress probability (**Figure 6**) [30].

Activities and their prizes produce strategies of the decision of an appropriate activity. In a given express Q work assesses the mean of the aggregate reward. "The perfect Q work is the greatest expected total reward that can be accomplished by following any of the arrangements". Reinforcement leaning is an ideal possibility for tending to different research difficulties in vehicular systems. For instance, helpful advancement of fuel utilization for a "given geographical region, taking care of the spatial and temporal varieties" of the V2V correspondences, ideal path expectation of electric vehicles, and decrease in traffic blockages.

#### **4.4 Deep learning technique**

"Deep learning is firmly identified with the above 3 classifications of ML". It is a more deeper system of neurons in different phases. It intends to extract information from the information expressions, that can be produced from the recently examined 3 classes of ML. The system comprises of input layer, hidden layers and an output layer. Every neuron has a non linear transformation function, such as ReLU, sigmoid, and flawed ReLU. The scaling of input information is critical as this can seriously influence the expectation or classification of a system. As the quantity of hidden layers builds, the capacity of the system to adapt additionally increments. However, after a specific point, any expansion in the hidden layers gives no improvement in the exhibition.

The training of a more deep system is additionally challenging since it requires broad computational resources, and the slopes of the systems may detonate or evaporate. The arrangement of these resources hungry more deeper systems has raised the significance of edge computing innovation. Vehicles progressing can get benefit from versatile edge computing servers. **Figure 7** illustrates the idea of neurons, input layer, hidden layers and output layer in a Deep neural network. It is also known as Multi-Layered Perception (MLP).

Deep learning intends to learn information portrayals, which can be worked in supervised, unsupervised, and reinforcement learning and has made huge advances in different machine learning assignments. As a deeper variant of neural systems,

which comprise various layers of neurons, the structure of deep learning is appeared in **Figure 7**. The input layer is at the left, where every hub in the figure speaks to a component of the information, while the output layer is at the right, relating to the outputs. The layers in the centre are called hidden layers [31].

#### **4.5 Unsupervised learning application in VANET**

Machine learning (ML) procedures are strong methods for classifying and clustering the enormous datasets because of their particular capacities, for example, short processing time, supporting the large amount of information, automatically detecting pattern in data, anticipating future data utilizing the uncovered patterns, and intending to gather more information [32]. In this survey, the ML algorithms are utilized for clustering the huge informational set and multi-dimensional highlights space of VANETs. Generally, the machine learning algorithms can be divided into supervised (classification) and unsupervised (clustering) learning algorithms categories [33]. In this survey we have focused on clustering technique in VANET which is the part of unsupervised learning method.

The supervised algorithms that are utilized for labeled information need to utilize a training informational set for looking at the highlights. However, the unsupervised algorithms that are utilized for unlabelled information do not have to utilize the training informational set [34]. "In this survey, the unsupervised machine learning algorithms are utilized for information clustering because of assorted variety of messages over V2V and V2I correspondences and unlabelled information in the VANETs".

"In current survey, then K-implies procedure is utilized as clustering then information in VANETs. Then K-implies is one of the most well-known unsupervised learning procedures broadly utilized as the multi-dimensional highlights clustering. The K-implies is an adaptable, fast and basic learning procedure. Additionally, this procedure is effective as preparing long information. In the specified, the K-implies research procedure utilized in the paper is depicted in major particular".

#### *VANET: A Machine Learning Approach DOI: http://dx.doi.org/10.5772/intechopen.109349*

K-implies procedure was at first presented "by James Macqueen in 1967" [27], yet its smooth one of the better mainstream "unsupervised" clustering procedures because of its effortlessness, experimental achievement, proficiency, and simplicity of usage. K-implies clusters a set of information into k number of clusters dependent on their highlights. For every data, K-implies computes the Euclidean distance to all cluster centroids (cluster centers), and afterward chooses the small distance. The information has a place with a cluster that the distance to the centroid of that cluster is the less. At that point, the new centroid is determined for each cluster dependent on the average coordinate of all members from each cluster. At last, all the information are clustered asked on the new centroid. K-implies repeat this procedure until the individuals from each cluster do not move to different clusters any longer [35, 36].

Essentially, K-implies comprises of 3 fundamental advances: "(1) selecting existing centroids as k clusters; (2) computing squared Euclidean distance of every data to the centroids; and (3) computing then new centroids cluster to discover nearest centroids. Stages 2 or 3 ought to be rehashed unroll the cluster individuals no more change [27, 29]."

The underlying centroids as K-means can be picked by Forgy and Arbitrary techniques [37]. In any case, there is no assurance that, utilizing these strategies, K-implies combines [37, 38]. In this way, the researchers utilize different techniques for deciding the underlying centroids. "In this survey, the underlying centroids for k clusters are thought to be the main k messages got by RSUs."

"K-implies has 3 inputs sources including highlights, number of clusters and number of iterations Clustering" procedures characterize a set of articles dependent on recognized highlights, along these lines, significantly affect execution of "the clustering procedures. In K-implies, the highlights must be changed to the grater qualities. Ordinarily, there is no proficient system as deciding the looks". In fact, the highlights ought to be resolved explicitly as every issue dependent on the information about the area of issues [27]. In the survey, the highlights of K-implies are characterized dependent on the highlights of "information including the message size, validity of messages, span between vehicles and RSUs, type of information and direction of information sender".

"The quantity of clusters are the alternative input as K-implies clustering procedure. The best number of cluster as every issue may be characterized by executing the clustering procedure as various quantities of clusters [27]. In current survey, the quantity of clusters (k) for K-implies is captured by leading a set of fundamental simulations".

The clustering procedure is exited when a predefined merging is produced. That explains there are not any modifies in then clusters' individuals. However, if the merging is not received in the considerable duration, the clustering procedure must be exit after a predefined number of loops. Theoretically, K-implies does not fastly assemble notable for the large information sets [27]. "In current survey, the number of loops is pretended to be 100".

The multifaceted nature of K-implies relies upon then quantity of information in every datum set, the quantity of highlights, number of groups and cycles. In this way, legitimate beginning conditions can bring about a superior bunching [32].

#### **5. Outcome formulation in VANET with learning-based methods**

The rich wellsprings of information produced and put away in vehicular systems inspire an information driven methodology for decision making that is versatile to

arrange elements and strong to different disabilities. Machine learning speaks to a successful apparatus to fill such needs with demonstrated great execution in a wide assortment of utilizations.

#### **5.1 Network congestion management**

Information traffic congestion is a significant issue in vehicular systems, particularly when the system conditions are profoundly dense, e.g., occupied crossing points and swarmed urban situations. In such cases, more number vehicles are competing for the accessible correspondence channels at the same time and thus cause serious information impacts with expanded packet misfortune and delay. To ensure a dependable and convenient conveyance of different deferral sensitive safety basic messages, for example, BSMs, the vehicular systems need to have carefully structured procedure rate-based, power based, carrier-sense multiple access/collision avoidance based, prioritizing and scheduling-based, and hybrid strategies [39], which alter correspondences parameters, for example, transmission power, transmission rates, and dispute window sizes, and so forth, to meet the congestion control purposes (**Figure 8**).

Unique in relation to the traditional methodologies, a viable machine learning based on information traffic jam regulation procedure using k-implies clustering has been created in Ref. [29] for traffic jam inclined convergences. The expected methodology depends on nearby "road side units (RSUs)" introduced at every convergence for traffic jam discovery, data "purification, and traffic jam regulation to arrange a subjective traffic jam authority to every vehicles that are going concluded or close at the" crossing point. Afterwards discovery of traffic jam, all RSU gathers all information moved among vehicles in their inclusion, expels its repetition, misuses k-implies procedures into cluster then informations as indicated by its highlights, for example, content, gravity and groups, lastly changes interchanges parameters for each cluster.

#### **5.2 Security in network**

As clever vehicles become increasingly associated and carry tremendous advantages to the general public, the improved availability can make vehicles progressively vulnerable against digital physical attacks. Subsequently, security of data partaking in vehicles are significant afterward any broken sensor estimations may source of hazards and wounds. In this chapter "intrusion detection system" has suggested to vehicular systems dependent on profound "neural networks", point

**Figure 8.** *Intersection of congestion area and RSUs.*

#### *VANET: A Machine Learning Approach DOI: http://dx.doi.org/10.5772/intechopen.109349*

the unsupervised rooted profound conviction systems are utilized into introduce the confines as pre preparing phase. At that point, "the profound neural networks are prepared by large-spatial packet information to make sense of the hidden measurable properties of typical and disfigure packets" and concentrate the relating highlights. Moreover, LSTM is utilized in to recognize attacks on associated vehicles. The LSTM based locator can perceive the incorporated abnormalities with high accuracy by figuring out how to anticipate the following word beginning from every vehicle.

#### **5.3 Location forecasting based set up and routing**

We are appeared in Segment IV that machine learning may be utilized to become familiar with the elements in higher versatility vehicular systems, along with vehicle direction forecast. In reality, the anticipated elements can be additionally utilized close to systems contract structures for framework execution enhancement. For analysis, the "Hidden Markov model (HMM)" has been enforced in to anticipate vehicles' forthcoming areas dependent on previous portability "follows and development designs in a mixed VANET along with V2I and V2V" joins. In light of the anticipated vehicle directions, a successful routing plan has been proposed to proficiently choose hand-off hubs for message sending and empower consistent handoff somewhere in the range of V2V and V2I correspondences.

"A variable-request Markov model has been received to separate vehicular versatility designs from genuine follow information in a urban vehicular system condition, which is utilized to anticipate the potential directions of moving vehicles and create productive expectation based delicate steering conventions." "In Ref. [28], a recursive least squares procedures has been utilized for enormous scope channel forecast dependent on the location data of vehicles and encourage the improvement of a novel planning procedure for helpful information dispersal in VANETs."

#### **6. Conclusion and future scope**

In this chapter, we have explored the chance of applying machine learning to address issues in high versatility vehicular systems. Solid elements displayed by such kinds of systems and the requesting QoS necessities challenge the cutting-edge interchanges innovations. Machine learning is accepted to be a promising answer for this test because of its striking execution in different computer-based intelligence related dimensions. We have analyzed the fundamentals of machine learning and afterward gave a few instances of utilizing such tools to get familiar with the elements and perform intelligent and dynamic in vehicular systems. We have additionally featured some territories that require more consideration.

This article gives a review of applying machine learning to address difficulties in developing vehicular systems. We have focused the basics of machine learning, including significant classes and agent methods. In this work examples of some models which are applying in machine learning in vehicular systems to encourage information driven dynamic, and discussed intelligent wireless resource management.

We have discussed different artificial intelligence procedures; all advancement issues face vulnerability in the intensions of the encompassing members. Various parts of artificial intelligence can help each other to draw out an ideal arrangement that would not cause or produce issues in the area they are not planned.

The machine learning procedures as a rule require higher computational assets. These assets may not require being inside the vehicle.

To for all intents and purposes to implement the review system in real vehicular systems, a RSU should be set at every crossing point. Likewise, since congestion control is a continuous procedure, RSUs may should be outfitted with Graphic Processing Units (GPUs) for rapidly executing machine learning procedures. We have to remember that, the machine learning procedures direct an enormous number of procedures and activities that takes a great deal of time.

#### **Author details**

Omkar Pattnaik1 \*, Sasmita Pani1 \* and Binod Kumar Pattanayak<sup>2</sup> \*

1 Department of Computer Science and Engineering, Government College of Engineering, Keonjhar, Odisha, India

2 Department of Computer Science and Engineering, Institute of Technical Education and Research, Siksha 'O' Anusandhan Deemed to be University, Bhubaneswar, Odisha, India

\*Address all correspondence to: omkar29in@gmail.com; susmitap20@gmail.com and binodpattanayak@soa.ac.in

© 2023 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

### **References**

[1] Pattnaik O, Pattanayak BK. Security in vehicular ad hoc network based on intrusion detection system. American Journal of Applied Sciences. 2014;**11**(2):337-346

[2] Pattnaik O, Pattanayak BK. Reliable and faster data communication in MANET and VANET based on localization and clustering techniques. International Journal of Engineering & Technology. 2018;**7**:66-72

[3] Liang L, Peng H, Li GY, Shen X. Vehicular communications: A physical layer perspective. IEEE Transactions on Vehicular Technology. 2017;**66**(12):10647-10659

[4] Peng H, Liang L, Shen X, Li GY. Vehicular communications: A network layer perspective. IEEE Transactions on Vehicular Technology. 2017. pp. 1-32 Available from: https://arxiv.org/ abs/1707.09972

[5] Cheng X, Yang L, Shen X. D2D for intelligent transportation systems : A feasibility study. IEEE Transactions on Intelligent Transportation Systems. 2015;**16**(4):1784-1793

[6] Cheng X, Fang L, Hong X, Yang L. Exploiting mobile big data: Sources, features, and applications. IEEE Network. 2017;**31**(1):72-79

[7] Jabbarpour Sattari M, Noor RM, Ghahremani S. Dynamic congestion control algorithm for vehicular ad hoc networks. International Journal of Software Engineering and Its Applications. 2013;**7**(3):95-108

[8] Liu Y, Dion F, Biswas S. Dedicated short-range wireless communications for intelligent transportation

system applications: State of the art. Transportation Research Record. 2005;**1910**:29-37

[9] Qian Y, Moayeri N. Design of secure and application oriented VANETs. In: Proceedings of IEEE VTC 2008 Spring. 2008. pp. 2794-2799

[10] Pedestrian Safety Plan. 2010. Available from: https://www.brampton. ca/EN/residents/Roads/pedestriandriver-safety/Documents/Pedestrian%20 Safety%20Plan%20-%20Final.pdf

[11] Taherkhani N, Pierre S. Improving dynamic and distributed congestion control in vehicular ad hoc networks. Ad Hoc Networks. 2015;**33**:112-125

[12] Grover J et al. Machine learning approach for multiple misbehavior detection in VANET. In: International Conference on Advances in Computing and Communications. Berlin, Heidelberg: A Springer-Verlag; 2011

[13] Bißmeyer N et al. Central misbehaviour evaluation for vanets based on mobility data plausibility. In: 9th ACM Workshop on Vehicular Inter-Networking, Systems, and Applications. Low Wood Bay, Lake District, UK; 2012. ACM 978-1-4503-1317

[14] Ismo K et al. Outlier detection using k-nearest neighbour graph. In: Proceedings of the 17th International Conference on Pattern Recognition. 2004

[15] Li W et al. A new intrusion detection system based on KNN classification algorithm in wireless sensor network. Journal of Electrical and Computer Engineering. 2014. pp. 1-8

[16] Maglaras LA. A novel distributed intrusion detection system for vehicular ad hoc networks. International Journal of Advanced Computer Science and Applications. 2015;**6**:4

[17] Nyakundi E. Using support vector machines in anomaly intrusion detection. PhD Thesis. 2015

[18] Sharafaldin I et al. Towards a reliable intrusion detection benchmark dataset. Software Networking. 2017;**1**:1

[19] Wang L. Support Vector Machines: Theory and Applications. Vol. 177. Berlin Heidelberg: Springer Science & Business Media; 2005

[20] Viriyasitavat W, Boban M, Tsai HM, Vasilakos A. Vehicular communications: Survey and challenges of channel and propagation models. IEEE Vehicular Technology Magazine. 2015;**10**(2):55-66

[21] Bernad'o L, Zemen T, Tufvesson F, Molisch AF, Mecklenbräuker CF. Delay and Doppler spreads of nonstationary vehicular channels for safety-relevant scenarios. IEEE Transactions on Vehicular Technology. 2014;**63**(1):82-93

[22] Guerrero-Ibáñez J, Flores-Cortés C, Zeadally S. Vehicular adhoc networks (VANETs): Architecture, protocols and applications. In: Next-Generation Wireless Technologies. London, U.K.: Springer-Verlag; 2013. pp. 49-70

[23] Zhang R, Cheng X, Yang L, Shen X, Jiao B. A novel centralized TDMA-based scheduling protocol for vehicular networks. IEEE Transactions on Intelligent Transportation Systems. 2015;**16**(1):411-416

[24] Elkan C. Results of the KDD'99 Classifier Learning. ACM SIGKDD Explorations Newsletter. 2000;**1**(2):63

[25] Li Y, Cimini LJ, Sollenberger NR. Robust channel estimation for OFDM systems with rapid dispersive fading channels. IEEE Transactions on Communications. 1998;**46**(7):902-915

[26] Li Y. Pilot-symbol-aided channel estimation for OFDM in wireless systems. IEEE Transactions on Vehicular Technology. 2000;**49**(4):1207-1215

[27] Liang L, Ye H, Li GY. Towards intelligent vehicular networks: A machine learning framework. 2019:1-12. arXiv: 1804.00338v3

[28] Tong W, Hussain A, Bo WX, Maharjan S. "Artificial intelligence for vehicle-to-everything: A survey", special section on emerging technologies on vehicle to everything (V2X). IEEE Access. 2019;**7**:10823-10843

[29] Teh YW. Dirichlet process. In: Encyclopedia of Machine Learning. New York, USA: Springer science+ Business media LLC; 2011. pp. 280-287

[30] Li Y, Cimini LJ. Bounds on the interchannel interference of OFDM in time-varying impairments. IEEE Transactions on Communications. 2001;**49**(3):401-404

[31] Ding W, Yang F, Dai W, Song J. Time-frequency joint sparse channel estimation for MIMO-OFDM systems. IEEE Communications Letters. 2015;**19**(1):58-61

[32] Sutton RS, Barto AG. Introduction to Reinforcement Learning. Vol. 135. Cambridge, MA: MIT Press; 1998

[33] Box GE, Tiao GC. Bayesian Inference in Statistical Analysis. Vol. 40. John Wiley & Sons, University of California; 1992

[34] LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;**521**(7553): 436-444

*VANET: A Machine Learning Approach DOI: http://dx.doi.org/10.5772/intechopen.109349*

[35] Basak D, Pal S, Patranabis DC. Support vector regression. Neural Information Processing. Letters and Reviews. 2007;**11**(10):203-224

[36] Gan G, Ma C, Wu J. Data Clustering: Theory, Algorithms, and Applications. Vol. 20. Philadelphia, PA, United States: Society for Industrial and Applied Mathematics, University City Science Center; 2007

[37] Araniti G, Campolo C, Condoluci M, Iera A, Molinaro A. LTE for vehicular networking: A survey. IEEE Communications Magazine. 2013;**51**(5):148-157

[38] Kenney JB. Dedicated short-range communications (DSRC) standards in the United States. Proceedings of the IEEE. 2011;**99**(7):1162-1182

[39] Abboud K, Zhuang W. Stochastic analysis of a single-hop communication link in vehicular ad hoc networks. IEEE Transactions on Intelligent Transportation Systems. 2014;**15**(5):2297-2307

#### **Chapter 6**
