Abstract:
The Internet of Vehicles (IoV) is a new paradigm of vehicular networks inspired by the adoption
of the Internet of Things (IoT) in Vehicular Ad-hoc Networks (VANETs). There are mainly two
types of applications used in IoV; safety, and non-safety applications. Because communication of
IoV is vehicle-to-anything; it is compromised by different types of attacks, such as denial-of –
service (DoS) attacks and impersonation attacks. DoS attacks create security problems in
vehicles by flooding unnecessary messages and creating congestion that leads to the safety
message being not delivered on time for the receiver vehicle.
Attackers can get the advantage of releasing fake or malicious data to the vehicular network by
controlling either onboard unit (OBU) or roadside unit (RSU). Previously published paper
deploys their deep learning classifier module either of OBU, edge server, or RSU. As they
employ on OBU of vehicular network attackers can penetrate anomalous data by controlling
RSU. When they deploy on an edge server there is a communication delay and need a high cost
for adding an edge server in different locations. Lastly, they deploy on trusted vehicle network
infrastructure RSU, but they don’t use any alternative in the case of RSU stop working. By
considering all the challenges of previous work, we are proposing anomaly based intrusion
detection system on semi-fog of IoV by using deep learning algorithms. The deep learning
algorithm is employed on RSU and OBU by selecting the vehicle as cluster head in case RSU
stops working.
In this work, we are using two deep neural network models, MLP (multilayer perceptron) and
LSTM (long short term memory), for training and testing a dataset. In this deep learning, we
train and test our deep neural network model by using WSN-DS and by customizing it to our
deep learning scenario. The dataset consists of an imbalanced dataset distribution and we are
applying SMOTE (synthetic minority oversampling technique) to balance the dataset. We are
training and testing the dataset after the SMOTE technique is applied to it to remove overfitting
problems. This paper is a novel one in terms of applying the deep learning classifier module on
the fog layer or RSU and OBU to a selected vehicle by selecting a cluster head in case
RSU stops working. After applying the SMOTE technique, we got 99.1%, 98.8%, 99.3%, and
99.6% accuracies for flooding, TDMA, grayhole and blackhole attack respectively. For
validating our deep learning model we use FC-BOT-IOT dataset and we got 99.70% accuracy
