Machine learning based novel frameworks developments and architectures for secured communication in VANETs for smart transportation

: Vehicle ad hoc networks have made intelligent transportation systems that significantly increase road safety as well as management possible. Vehicles can now communicate and share information about the road using this new technology. However, malicious users might inject fake emergency alerts into VANET, making it impossible for nodes to access accurate road information. In vehicular ad hoc networks, assessing credibility of nodes has become a crucial task to ensure reliability as well as trustworthiness of data. Using machine learning methods, this study proposes a novel security technique that improves communication and intruder detection in VANET for smart transportation. Ciphertext-policy game theory encryption analysis for smart transportation is used here to improve the security of the VANET. Fuzzy rule-based encoder perceptron neural networks are utilized in the detection of the VANET intruder. For a variety of network datasets, the experimental analysis is conducted in terms of throughput, QoS, latency, computational cost, and data transmission rate.


Introduction:
A mobile network known as VANET allows vehicles to communicate with one another in absence of fixed infrastructure.Goal of this type of network is to improve road safety by exchanging alerts between nearby vehicles or to provide new comfort services to road users [1].Unfortunately, ad-hoc network vulnerabilities are not limited to issue of shared wireless medium but also include utilized routing mechanism as well as auto-configuration.The trust that exists between the participating nodes underpins these mechanisms.All cooperative network services will be disrupted if a node engages in malicious behavior [2].The implementation of an IDS is an efficient method for determining when an attack occurs in a VANET.An intrusion detection system (IDS) is a tool for spotting suspicious or abnormal activity on a target (a network or host).It makes it possible to know whether or not an intrusion attempt was successful.Internal attacks can be detected with the help of IDS solutions.These are attacks that cannot be detected by cryptographic methods.Indeed, compromised nodes carry out internal attacks.After the cryptographic systems, an IDS is frequently utilized as a second line of defense.An intrusion detection system typically includes three phases: one for data collection, one for analysis, and one for responding to an attack to stop it or lessen its effects on the system.IDS is housed on some specialised nodes known as monitors or monitoring nodes.IDS's protocol type and architecture both influence how these nodes are set up [3].
On the Intelligent Transportation System (ITS), VANETs can be utilized to guarantee road users' safety and comfort.The primary goal of VANETs is to safeguard the vehicle, drivers, and passengers.By exchanging cooperative awareness messages (CAMs), these networks provide security as well as accomplish their objectives.In addition, self-driving cars rely heavily on these networks because they free up space and time and provide the vehicle with information when it is needed.However, these networks have vulnerabilities at every level due to certain characteristics.These attributes are: absence of traditional security infrastructure, highly dynamic network topology, and open wireless medium.Their vulnerability has also been made worse by other issues, such as the auto-configuration feature and the routing mechanism [4].In comparison to conventional networks, VANETs are distinguished by their high speed, mobility, and vehicle density.Purchasing network security is difficult and risky due to these features [5].Self-driving vehicles are a new type of vehicle currently under investigation.Because they did not officially become vehicles until just recently, these vehicles are regarded as a revolution in automotive industry.Prior to the introduction of self-driving vehicles, communication companies and vehicle manufacturers worked together.On Board Units (OBUs), which include unidirectional antennas, at least one processor, a Global Positioning System (GPS), and a collection of sensors for V2V and V2I communications, have begun to be included in automobiles [6].

Related works
The study of the literature was done by looking at a number of papers that are based on IDS in the internet using a ML method and a variety of algorithms that are used for internet security.Trust based method with a classification strategy employing Bayesian NN (BNN) was proposed by Elvin Eziama.This model uses a classification approach and is generic in nature.Numerous researchers have proposed a variety of algorithms, including trust-based algorithms [7] and cryptographic solutions [8].If the attacker is an insider, trust-based solutions are useful.The cryptographic solution is effective against both internal and external attackers.Large calculations incur some overhead in cryptographic solutions, which may delay message delivery.The primary disadvantage of cryptographic solutions is this.A robust method for detecting malicious vehicles for the Post Crash Notification application has been proposed in the study [10].They have thought about the possibility of a false crash alert in [11] and fake vehicle position data in PCN.A novel Misbehavior Based Reputation Management Scheme (MBRMS) for detection as well as filtering of false data in vehicular ad-hoc networks has been proposed by work [12].It consists of 3 parts: misbehavior detection, event rebroadcast, and global eviction methods.A detection algorithm known as DMV has been proposed by work [13] to distinguish honest nodes from malicious nodes by duplicating or dropping received packets.The allocated verifier nodes keep an eye on the vehicles that have been tagged with a distrust value.Author [14] used a cooperative watchdog method based on the Dempster-Shafer model to detect malicious vehicles in the VANET utilizing Quality of ServiceOptimized Link State Routing (QoS-OLSR) clustering method.With an increase in detection probability as well as decrease in number of selfish nodes as well as false negatives, this strategy maintains service stability and quality.A novel strategy for not only detecting malicious vehicle attacks but also preventing them from the VANET has been presented in work [15].Detection of Malicious Vehicles (DMV) method has been improved.Impact of a black hole attack on VANET was lessened by this strategy, which is also more effective and secure than DMV.Based on secondary data or alerts that are generated in response to primary alerts for PCN application, researchers [16] have detected misbehavior in VANETs.Concept of a data-centric misbehavior detection method program was utilized in a brand-new misbehavior detection scheme that was presented by [17].[18] devised a VARM mechanism that gathers information from neighbor transmissions at a single vehicle in order to locate the malicious vehicle.Work [19] has proposed a novel algorithm for probabilistic driver detection to secure communication in VANET.It determines whether message is from an honest vehicle and calculates the trustworthiness of the messages received.Using a machine learning technique, work [20] presented a security framework for classifying numerous VANET misbehaviors.Based on characteristics determined by observer nodes, it distinguishes between a malicious node as well as honest node.

Materials and methods:
In this section, a machine learning-based, smart transportation-based method is proposed for secure communication and malicious activity detection.The edge device takes place of the road side unit in proposed approach.As a result, when edge device receives a request for service from one of vehicles, it responds immediately as well as sends summary of process that was carried out independently to the cloud service.Figure 1 depicts the VANET alignment architecture.

Figure-1 VANET alignment architecture
Stages of security assistance as well as performance enhancement offered to vehicular network by edge and cloud computing are described in the following step.V1 performs a registration or authentication verification process upon receiving an alert from V2 and RSU to determine whether the message came from an authentic source or a malicious node.Figure 2 shows that the mode in which vehicles communicate is called ad hoc, and that mode changes to infrastructure mode when infrastructure is added.Step1: Communication is carried out over wireless LAN using edge device rather than the RSU for a network (N) with a significant number of vehicles (V), whereV = "V1,V2,..., Vi," in order to overcome delay in data transmission and have a quick process.
Step 2: In order to avoid unnecessary hacking and attacks, approximate member query filter (AMQF) is used for security purposes in both vehicles and edge.Each vehicle in network keeps an AMQF of all vehicles in network.When a vehicle enters arena of an edge node, edge node updates data in vehicle list table that is maintained, and enrollment includes vehicle's identification number (Vid).In response, edge node provides a common public key for all vehicles in edge arena.
Step3: AMQF is used to determine whether the vehicle requesting an interaction with an edge or vehicle is an attacker or a member.After it is determined to be a member, acknowledgement process begins, paving the way for transmission and reception of data.otherwise closes communication after checking with update table to see if it is a member in the edge case of V2V communication.This is not required for vehicle-to-edge communication because the edge consults update table prior to interacting with any communicator who requests a service.

Ciphertext-policy game theory encryption analysis for smart transportation:
Five main components of our structure are as follows: Setup(U): System's attributes set U and security parameter are used as inputs in the setup step, and public key and master secret key are used as outputs.Finally, public key PK is given as eq.( 1) Master secret key is set to be MK = gα.
Encrypt(PK,(M,ρ),m): Method will definitely use a message m as input during the encryption step.In addition, the message should be encrypted using public key PK and an access structure before corresponding ciphertext is produced.Method first selects a random vector  = (, )  ,  ′ =   ,, in which exponent s ∈ Zp is randomly selected as secret to be shared.For i = 1 to l, it evaluates λi = Mi v. Additionally, it should select several random exponents r1,...,rl ∈ Zp for encryption calculation.Ciphertext CT is given as eq.( 2) Generate a key (MK,S): Method uses master secret key and a set of user characteristics as input during key generation phase.The algorithm begins by selecting t ∈ Zp, a random number.The private key is created in two parts: AK, a "attribute key" for proxy, and SK, a private "security key" for user's equivalent (3).
Transform Ciphertext(CT,AK): This technique accepts as inputs an attribute key AK for a set of attributes S and a ciphertext CT which designates a valid user.We define {  ∈ ℤ  } ∈ as a collection of constants and represent I ⊂ {1, 2,...,l} as I = {i,ρ(i) ∈ S}.Only if ∑ ∈     = holds, in accordance with M's access structure, can we assert that λi are legitimate shares of secret s.
1. Proxy determines if users can solve the following equation to decrypt the ciphertext (4).
2. The ciphertext CT is converted by the proxy into altered ciphertext CT' by eq (5).
Decrypt(CT',SK): This method receives the user's private key SK and transformed ciphertext CT' as inputs, and then it outputs plaintext m via eq (6).
(SK,  ′ )CT ′ = (  2   ,   )(, )  1  /(, )  = (, )  2  (, )  (, )  1  / (, )  = (, )  (6) m = C/e(g, g)s will result in the plaintext m.One pairing computation is all that is required for this kind of decryption.The plan not only ensures the safety of data, but it also speeds up user decryption.The cluster is formed using the game theory method.System is broken up into 3 main parts, which are listed below, to form the cluster: (1) Vehicle of Origin: It is the medium through which messages and data are transmitted in accordance with an individual's requirements.
(2) The Hop Car: +These vehicles serve as a system for forwarding data.It is helpful to send the data as soon as possible to the destination.
(3) Point of Access: The VANET's stationary infrastructures serve as access points.
The cluster's systematic diagram given in Figure 3.In order to transmit the data to access point and other nearby hop vehicles, each cluster has one source vehicle as well as one additional hop vehicle.With the aforementioned data, there will be X source vehicles and Y hop vehicles for the game to play.We are going to assume, for the sake of simplicity, that the source vehicle will transmit data using a single hop vehicle.Game theory method for selecting CH is depicted in Table 1.VANET is first put into operation using this game theory strategy, and then it is classified as either a source vehicle or a hop vehicle.Since each vehicle's social parameters are now identified, this approach to the VANET analysis is more novel.The CH selection is made using K-mean clustering after the social scores have been calculated.Game theory is utilized to select best CH from available options after choosing a potential CH.
The proposed method stands out from the other traditional approaches using this strategy.
For each element, it is simple to establish type-II fuzzy sets.Equation is a type II fuzzy set's definition ( 5) where µL and µU stand for the initial membership function (x), defined as eq(6) lower ,'s and upper membership degrees, respectively where α ∈ (1,∞).The sigmoid activation function is one that is frequently utilized in multilayer perceptrons as an activation function.Sigmoid function introduces non-linearity in hidden layers, allowing neural network to learn more intricate information.
Type-II fuzzy sigmoid activation function ϕ is expressed as eq (8) consider that is fuzzy sigmoid activation function where, ϕL and ϕU represent lower and upper sigmoid activation functions.Equation ( 9) defines suggested fuzzy gradient descent.
where w denotes weights, u1 and u2 denote the degrees to which every neuron belongs to class 1 and class 2. A hyperplane would be modified to fit a training dataset when using the perceptron for supervised linear classification tasks.The classification of new, unknown samples can then be done using this tuned hyperplane.This is accomplished by minimising the error function while the hyperplane is applied to the training dataset: () = − ∑ i∈      x i , classes are entirely separated by the hyperplane.Typically, this minimising procedure is carried out in iterations, with every iteration leading us closer to the minimum of ǫ(w).Weight updating step wk+1 = wk + 1w yields w vector of iteration k + 1. (weight update).Learning rule utilized to get value for updating weights at every increment is shown in equation (10): where truej is actual class label, predj is anticipated class label, and η learning rate.To start the learning process, the perceptron's weights are initialised to small random integers [or 0].Up until a minimum error is reached, weights are changed and the output value is calculated for each training input sample (backpropagation).To maximise model prediction, gradient descent, which is defined as eq.11), can be used to locate local minimum of a function and reduce network error.
where the learning rate is η, w are weight values, and   ( is derivative of sloperepresenting objective function F(w).
Think of a binary classification issue where examples are z = (x, y) ∈ R d × {−1,+1}.Primary cost function as eq is minimised to get the linear Q-learning classifier.( 12) |  ⟩ = (cos   , sin  ˙)  and   = arctan (∑ =1  sin(  +   )/∑ =1  cos(  +   )) then equation ( 11) is given as eq.( 13) We can convert our discrete time Markov Chain into a continuous time differential eqn using the expansion ( 14) where we have defined eq. ( 15) Then, we add the aforementioned phrase to Π   , which results in eq. ( 16): The associated differential equations have the form eq(17) using the aforementioned definitions where M gives the tensor M The aforementioned system of differential equations can be reduced as eq (18) Where we defined ℎ =    gh ⊗  × with: eq. ( 19) After acquiring the temporal features of a frame at time t, we receive a binary output with the objects as one class and background as another class.VOP is created from the pixels that correspond to the FMt section of the original frame, and the moving object areas are identified as the foreground regions in the temporal segmentation.
The suggested model's numerous flag values are used to characterise the algorithm.In order to minimise communication overhead, we solely use binary trust values to determine a node's level of trust.There is only a 1 bit overhead difference when transferring trust data to other network nodes.Tags (Boolean flags) utilised in suggested method are defined as follows: T (AB) is the node A-calculated combined trust of node B, and T (B) is the node A-calculated direct trust of node B. T is a representation of the suggested trust (the total of the other nodes' trusts) (P Bi).Node movement in a single lane is uniform.To track the behaviour for as long as eq (20), the observed nude and the observer node must face the same way.
where  ()  >  denotes the direct trust of node B and m denotes number of modified packets, d number of discarded packets, f the number of forwarded packets without modification.We set flag value for direct trust based on direct trust value mentioned above.We determine     The primary objective of proposed system is to identify malicious vehicles in VANETs and implement secure communication between self-driving cars.Three phases are used to implement this system: phases of data collection and preprocessing, training, and testing.In addition to making a comparison with anticipated theoretical results to guarantee system's effectiveness, this method is distinctive in that it combines a variety of methods to achieve the best results.However, main issue with anomaly detection is high rate of false positives due to various records that describe both normal as well as abnormal behavior.VANET ultimate objective is to give vehicles with comfort as well as safety without jeopardizing their security or privacy.Modifying or employing an effective algorithm that ensures the safe transfer of information between vehicles is needed to maintain the security of vehicular network.A trustbased method can be utilized to find malicious network nodes.Without increasing network's control overhead, malicious nodes are eliminated from network using our proposed model.We choose the group of observer nodes to monitor network node.The status of a particular node is communicated in form of a binary flag by observer node, which also calculates combined trust of that node.Due to utilization of observer node, load is evenly distributed to all of nodes, and lightweight nature of our proposed model means that it requires less computation.Without much overhead, proposed model facilitates safe vehicle communication.Range of malicious activities that our proposed model can detect is limited.The proposed model can only be used to find internal attackers.

Conclusion:
Using ciphertext-policy game theory encryption analysis for smart transportation and fuzzy rule-based encoder perceptron neural networks, this study proposes a novel method for security-based network communication with intruder detection.Our method divides vehicles into clusters, each of which is managed by a roadside unit.Before gaining access to vehicular ad hoc networks, RSU checks credibility of nodes.A malicious node is removed by the roadside unit based on its trust value.Our method is validated through simulations.By lowering ratio of malicious nodes in VANET, we demonstrate that our strategy is able to detect and remove all malicious nodes over time.Throughput, quality of service, latency, computational cost, and data transmission rate were all achieved using the proposed method.In the future, we intend to make use of a verification tool to guarantee that our plan is resistant to a variety of attacks as well as that it only chooses VANET nodes with highest level of trust.In addition, we would like to carry out a similar analysis on a system in the real world.

Figure 3 :
Figure 3: Systematic diagram of cluster

Figure- 5
Figure-5 flow chart of proposed VANET security

Table 1 :
Game theory-based CH selection