Document Type : Research Paper
Authors
1 Assistant Professor, Department of Computer Science & Engineering at SRMSCET, Bareilly (UP) India, Affiliated to Dr. A.P.J. Abdul Kalam Technical University, Lucknow, Uttar Pradesh, India.
2 Professor and Dean Academic Affairs, Sharda University Greater Noida (U.P.) India, Pin Code:201306.
3 Professor and Director, KEC, Ghaziabad (U.P.) India, Affiliated to Dr. A.P.J. Abdul Kalam Technical University, Lucknow, Uttar Pradesh, India.
Abstract
Keywords
(VANET)Vehicular Ad-hoc Network is a subset of (MANET) Mobile Ad-hoc Network), having frequently changing topology, high mobility and a few other intricacies (Bakhouya et al., 2011). VANETs are susceptible to big range of security threads due to the openness of network, lacking of fixed infrastructure, in which the participating nodes (vehicles) can create a network freely with no requirement of pre-deployed communication framework (Ali et al. 2020). The structure of VANET is specified is given below:
Wormhole attack is such type of attack which can interrupt the routing mechanism of AODV and DSR routing protocol in VANET (Albouq & Fredericks, 2017). The wormhole contains at least two malicious vehicles. These malicious vehicles form a tunnel (Singh et al., 2019). Following figure 2 shows a wormhole assault:
Figure 2.Wormhole assault (Upadhyaya & Shah, 2018)
The intention of these malicious vehicles is to mislead permissible vehicles and pretend them as they are neighbors (Singh et al., 2019). The present security structure, based on public key encryption is invulnerable to outsider assaults. To detect such type of misbehaving nodes inside the network is very difficult (Singh et al., 2019). In this research paper, a machine learning based approach is used, to detect the wormhole assault in VANET. A scheme which is combination of packet leash (Ali et al., 2017) and cryptography is also proposed to stave off the wormhole assault. Here, research work depends on wormhole assault with Ad-hoc On-Demand Distance Vector (AODV) routing protocol. First, machine learning models are trained, and then these machine learning models are used to foretell the deportment of every vehicle in VANET according to rules those machine learning paradigm have learned. In this research work the performance of proposed ML models is also compared with existing work.
In (Bakhouya et al., 2011), the authors given an adjustive approach for information metastasizing in VANETs. In this method, every participating node stabilizes the values of local parameters utilizing local data (such as number of superfluous messages, inter-arrival time), related to the messages which are obtained from neighboring vehicles with no attempt.
Strengths:
Weaknesses:
Redundant broadcast messages indirectly affect the performance of network.
In (Albouq & Fredericks, 2017), the authors introduced a lightweight wormhole protocol-detector (WPD) protocol to discover as well as alleviate wormhole assaults.
Strengths:
Weaknesses:
In (Singh et al., 2019 ), a scheme is proposed based on ML to detect wormhole attack in VANETs.
Strengths:
Weakness:
In this research paper it is not discussed how the data is pre-processed.
(Hu et al., 2003) presented wormhole assault as well as cited how dangerous this assault is in MANETs. It's not easy to protect towards wormhole assault, and can form a big threat particularly towards several ad-hoc routing protocols.
Strengths:
Weaknesses:
In this proposed approach there is a problem of time synchronization node movement with speed of light.
In (Grover et al., 2011), a ML-based approach is discussed which is used to differentiate the vehicle’s behaviour.
Strengths:
Weakness:
Here WEKA tool is used, which cannot handle a big data set.
In (Kang & Kang, 2016), the authors suggested a (DNN) deep neural network supported process to find out.
Strength:
The proposed technique provides the real-time response for assault.
Weakness:
The proposed approach does not give the efficient result if data set is very big.
In (Loukas et al., 2018), authors suggested IDS based on deep learning, which is implemented on cloud to find cyber-physical assaults within the vehicle.
Strengths:
Weaknesses:
(Ali et al., 2016) proposed ML-based approach for detecting gray hole as well as rushing assault in vehicular plan.
Strengths:
Weaknesses:
At different plane, different machine learning based methods have been implemented to find out the assaults. But machine learning based models, to detect the wormhole assault in VANET over real map are used in few research works which do not have good detection performance. In this research work machine learning based approach is used to find out the wormhole assault in VANET with better performance as compare to existing work. So this work will have a significant role in research.
VANET is latest technology by which the journey can be made more easy and secure. In VANET participating nodes are the vehicles. Due to mobility of vehicles the topology changed very rapidly. So routing is main issue. Due to wireless nature of vehicles the VANET is more suspected to the attacks. The VANET can be made more secure by detecting and preventing various attacks such as wormhole attack. In this research paper ML based approach is used to detect the wormhole assault to make the VANET more secure. An approach based on cryptographic concept and packet leash is also proposed to stave off the wormhole assault.
VANET is a subset of MANET, in which V2V & V2I communication takes place to enhance traffic and security applications. Real map scenario is considered to detect the wormhole assault in VANET by using machine learning based approach. For the same VANET scenario, an approach based on cryptographic technique and packet leash is also given to prevent the wormhole attack.
In this work for V2V and V2I correspondence, AODV routing protocol is considered. By proposed approach of detection & prevention of wormhole attack, VANET can be made more secure in future.
Here, a machine learning based wormhole assault discovery system is proposed which find out the conduct of participating vehicles in vehicular ad-hoc networks. The approach utilizes the data collected from simulator which contains both normal as well as abnormal conduct (under wormhole attack) of vehicles in VANETs.
The steps which are followed in this proposed approach are given as follow:
The scenario for VANET is given as follow:
Figure 3. Scenario for VANET
K-nearest neighbors (k-NN) and Random Forest machine learning models are used in this research work.
The k-NN algorithm (Altman, 1992) may also resolve each categorization & retrogression issues, but primarily being applied for categorization work. For each data point, we have assumed k closest training datums as well as predict the most happening class for any test datum. Here k is a hyper parameter which expresses training datum numbers to be assumed for labelling test datum.
Definitely Random Forest is a group classifier, which creates a group of self-sufficient as well as non-identical decision trees according to the idea of randomization. Random forest may be defined as {h(x,θk),k=1,2,……L}, in which θk is a kind of mutual self-sufficient random vector variables, as well as x is the input data (Provost et al., 2016). The random forest having following advantages:
An attacker can access to global parameter n and g which isn't having adequate information to possibly calculate Ks. However, an assault might be able to intercept and modify all messages as well as negotiate a secret key KA with A and KB with B. The standard Diffie-Hellman key distribution scheme is prone to such man-in-the middle attacks.
The proposed scheme to prevent wormhole attack is divided into following two algorithms:
Proposed algorithm to distribute the shared key Ks
Algorithm proposed, to stave off wormhole assault in VANET
In proposed algorithm to distribute the shared key Ks, man-in-middle attack is eliminated by keeping parameters n and g secret; means the encrypted form of n and g tend to be pre-initialized in vehicle OBU’s memory before coming on the road. In case any vehicle is captured, hacker will be unable to read n and g.
Phase 1: (Before coming the vehicle on road): Design& deploy the same random number generator to every vehicle before coming on road. Assign the secret number S and the common number n and g to each vehicle manually, before using the vehicle on road in an encrypted form (n Å Ri && g Å Ri), in the memory of OBU’s, where:
Random Number GeneratorGenerator |
S→ →Ri
Phase 2 (key generation):
(a).Secret number X’ = Q Å S = q
(b).Decrypt the common number n and g.
The above algorithm (Kumar & Singh, 2016) is proposed to distribute the shared key in WSN. But in this research paper it is proposed for VANET. The above shared key (Ks) is used for the further communication to prevent the wormhole attack.
Figure 4: Flowchart of proposed approach
(dsr) distance between sender and receiver as follow:
dsr≤|ps-pr| +2v(tr-ts+∆) +δ.
in which
±∆=time drifting of sender and receiver
v=maximum speed of any vehicle
δ= the maximum error in location information
te=ts+dsr/c.
where:
c=wireless signal's propagation speed in VANET.
then packet is secured
else
if(tr≥te)
then packet is not secure and discard it.
Above algorithm is proposed in (Ali et al., 2017) research paper. But in (Ali et al., 2017) the key is distributed by using public key crypto system like RSA which requires more computations.
A dataset is created from simulation to learn the assault. The screenshot of simulation is given below:
Animation of scenario on NetAnim is given as follow:
Various features which are given by flow monitor tool of NS-3.24.1 simulator are source and destination IP, timeFirstTxPacket, timeFirstRxPacket, timeLastTxPacket, timeLastRxPacket, delaySum, jitterSum, lastDelay, txBytes, rxBytes, txPackets, rxPackets, and lostPackets etc. But in this research work txPackets(Transmitted Packets), rxPackets(Received Packets), lostPackets(lost Packets) and jitterSum(Jitter Sum) features are considered to create the dataset. The AODV routing protocol is modified to create the scenario for wormhole assault in NS-3.24.1for VANET.
The dataset which is created from data obtained from flow monitor of animator of NS-3.24.1 simulator is not in normalized form. To apply the ML model the normalized dataset is required. The function in python to normalize the dataset is given as follow:
normalized_df=(df-df.min())/(df.max()-df.min())
After normalization, the dataset is fetched in the form of excel sheet. The function in python to fetch the dataset is given as follow:
normalized_df.to_excel(writer, sheet_name='Sheet1')
(SUMO-0.32.0) Simulation of Urban Mobility Model (Khan et al., 2019) and (NS-3.24.1) Network Simulator (R.Henderson et al., 2008) are used as simulators in this research work. We have taken Noida (Near Sector 93A Park) (U.P.) India map with the help of OpenStreetMap (OSM) to generate various files.
The VANET scenario generated by the SUMO-03.2.0 with the help of Real Map is given below:
The steps used to generate various files are given in the following figure:
Various steps which are followed to create scenario are shown as follow:
Figure 9. The snapshot of mobility.tcl file is shown as follow:
The snapshot of VANET scenario file is shown as follow:
Figure 10. Screenshot of VANET Scenario file
Various parameters used in simulation are given as follow:
Table1. Parameters & Values used in Simulation
Parameter |
Value |
Simulator |
NS-3.24.1 |
Time of Simulation |
30 s |
Vehicles Strength |
31 |
Type of Traffic |
(CBR)Constant Bit Rate |
Territory Dimension |
5198.32m ∗ 3558.2 m |
Transport Protocol |
512 Bytes |
Transport Protocol |
UDP |
Protocol for Routing |
AODV |
MAC Protocol |
IEEE 802.11p |
Radio Propagation Model |
Two ray ground |
Maximum Speed |
20 m/s |
Mobility Model |
Noida (Near Sector 93A Park) Model |
For generating the dataset, simulation is carried out multiple times with a distinctive couple of vehicles as assaultive vehicles. We mark the overall transmission taking place between the pair of assaultive vehicles as abnormal and the remaining as normal.
Table 2. Attributes of Dataset
Variable |
Value |
Count of Features |
4 |
Categories |
2 (For Normal 0, for Abnormal 1) |
Total Cases |
249 |
0-Tag |
212 |
1-Tag |
37 |
Size of Training |
174 |
Size of Testing |
75 |
Discussion regarding the results which are achieved after the simulation, training and testing a machine from created data set is carried out, in this section. The data set is created by collecting the required data values from the flow monitor of animator. The screenshot of flow monitor flow id and normalized dataset is given as follow:
Figure 11(a). Screenshot of flow monitor with flow Ids
Figure 11(b). Screenshot of normalized dataset
The dataset is splitted into 70% training and 30% testing dataset. In python following syntax is used, first:
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
After importing the data set, the partitioning is carried out by using the following code:
From sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test=train_test_split(X, y,test_size=30)
The graph of training dataset and testing dataset is given as follow by using python programming on jupyter Notebook.
Number of training datasets |
Figure 12(a). Graph for training dataset for various features.
Number of testing datasets |
Figure 12(b). Graph of testing dataset for various features.
In this research work the value 3 is assigned to k in k-NN. The results show that the k-NN having the accuracy of 99.196% and random forest is having the accuracy 98.666% respectively. The screenshot of execution is given as follow:
Figure 13. Screenshot of results computation
In existing research paper (Singh et al., 2019) the k-NN model is having the accuracy 99% to detect the wormhole attack, but in our research work the k-NN model is having the accuracy 99.196% to detect the wormhole attack in VANET due to proper normalization of data. In existing research random forest ML model is not used to detect the wormhole attack in VANET. In this research work the random forest ML model is also used due to its various advantages as compare to other ML models as discussed previously. The random forest ML model also performed well with a detection accuracy of 98.666%.
And by analytically it is proved that the scheme based on packet leash and cryptographic technique is useful to prevent the wormhole attack in vehicular ad-hoc network.
In this research work an approach based on ML concept is proposed to detect the wormhole assault in VANET. By using methodology & results it has been demonstrated that how efficient ML is in finding wormhole assault in VANET. The depicted model finds the assault with accuracy 99.196% and 98.666%. ML can be utilized at various layers of protocol stack of VANET to detect the wormhole assault. An approach based on packet leash and cryptographic concept, is also proposed to stave off the wormhole assault in vehicular ad-hoc network.
Conflict of interest
The authors declare no potential conflict of interest regarding the publication of this work. In addition, the ethical issues including plagiarism, informed consent, misconduct, data fabrication and, or falsification, double publication and, or submission, and redundancy have been completely witnessed by the authors.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article