TM – IoV : A Dataset for the Trust Management in the Internet of Vehicles

1. Introduction and Background

Over the past decade or so, the rapid evolution and advancements in a number of cutting-edge technologies, including but not limited to, the Internet of Things (IoT), artificial intelligence, and the fifth-generation communication has led to the transformation of the conventional intelligent transportation systems into Internet of Vehicles (IoV) networks [1,2]. The IoV networks facilitate seamless connectivity for a real-time exchange of safety-critical and non-safety information amongst the vehicles, and between the vehicles and the vulnerable pedestrians, supporting roadside infrastructure, and the backbone network via vehicle-to-everything communication [3,4]. Despite the low latency advantages associated with the IoV networks, they are prone to a number of malicious attacks that are not only capable of jeopardizing the entire network but also poses a considerable risk to the human lives. Hence, it is of paramount importance to ensure the resilience of the IoV networks [5,6]. Figure 1 portrays an IoV landscape.

A brief glimpse of the state-of-the-art reveals that a number of mechanisms have been proposed over the years in order to strengthen the security of an IoV network. Such mechanisms can be broadly classified into two categories, i.e., cryptography-based approaches and trust-based approaches [7,8]. Whilst the cryptography-based approaches safeguard the IoV networks against a number of malicious attacks, including but not limited to, data tampering, identity theft, and eavesdropping, they are prone to a number of internal attacks [9,10]. Trust-based approaches, on the contrary, can intelligently address the challenges pertinent to internal attacks [11] since they leverage the reputation of entities within an IoV network in order to guarantee secure communication amongst them, thereby facilitating intelligent traffic flows [12,13].

In the context of an IoV network, vehicles are classified as either trusted or untrusted [14,15]. Trusted vehicles exhibit legitimate behavior by primarily disseminating accurate information in an IoV network, whereas, untrusted vehicles engage in malicious activities by intentionally transmitting and facilitating the relay of incorrect information and recommendations in an IoV network in an intelligent manner, thereby posing a grave threat to the vehicular passengers and the vulnerable pedestrians [16,17]. Hence, an accurate and real-time identification of malicious vehicles in such a highly dynamic network is highly indispensable [18]. The state-of-the-art methodologies employed for the identification of such vehicles in an IoV network typically involve threshold-based and decision boundary-based mechanisms [19,20]. In the case of threshold-based mechanisms, a vehicle’s trust value is compared with a predetermined trust threshold, i.e., if the trust value of a vehicle exceeds the predetermined trust threshold, it is regarded as a trusted vehicle, or else, it is classified as a malicious vehicle. On the contrary, in case of decision boundary-based mechanisms, the trust values derived from vehicular interactions are clustered and classified via learning algorithms, and an optimal decision boundary is subsequently employed to segregate the trusted vehicles from the malicious ones [21,22]. However, regardless of the methodology employed for the identification of the malicious vehicles in an IoV network, vehicles should have an associated precise trust value. Therefore, trust-related data holds a considerable significance for securing highly dynamic IoV networks.

Trust, in essence, implies degree of a belief or a disbelief that a trustor has on a trustee for carrying out a particular task or a set of tasks in an anticipated manner [23]. It mandates quantification, and to realize the same, it relies on several trust-based parameters which are not only context dependent but are also highly dynamic in nature since they transpire as a result of the frequent interactions amongst the vehicles in an IoV network [24,25,26]. Whilst a number of IoV-based trust parameters have already been delineated in the research literature, as of date, there is no dedicated publicly available trust-based IoV dataset that researchers from both academia and industry can predominantly employ in order to carry out an in-depth research and subsequently expand upon within this particular domain. In order to address this particular challenge, the manuscript-at-hand presents a pioneering trust-based IoV dataset which is discussed in detail in the Section 2 (Data Description) and Section 3 (Methods).

2. Data Description

The manuscript-at-hand introduces a trust-based IoV dataset which has been made available for the readers at https://github.com/wangyingxun/IoV. This particular dataset has been employed for not only ascertaining the trust values of vehicles in an IoV network but also for segregating the trustworthy vehicles from the untrustworthy ones by means of an optimal decision boundary. Accordingly, a detailed description of the key features of this particular dataset is indispensable so as to enable researchers in both academia and industry to employ and extend the same in a bid to investigate open research directions of this emerging and promising domain.

It is pertinent to mention here that to date, there is no public dataset pertinent to trust management in IoV networks. Therefore, the dataset proposed in the manuscript-at-hand represents a pioneering contribution within this particular domain. In order to realize the same, Java has been employed for designing an IoV-based simulator, whereas, Python was employed for analyzing the simulation results. The IoV simulator takes into account several interconnected road segments in order to mimic a road network encompassing vehicles traversing at random speeds in disparate directions. Vehicles, accordingly, interact with one another and frequently exchange indispensable information to realize a number of both safety-critical and non-safety applications. Moreover, the proposed IoV-based simulator incorporates not only honest vehicles but also intelligent malicious ones that dynamically alternate between honest and dishonest behaviors in a bid to execute malicious acts so as to evade detection by the IoV-based trust models [27,28].

For reference of the readers, the trust-based IoV dataset proposed in the manuscript-at-hand encompasses 79 vehicles, i.e., trustors and trustees, that engage in a total of $96,707$ interactions over different time instances. In total, we ascertained 9 key trust parameters, i.e., packet delivery ratio, similarity, external similarity, internal similarity, familiarity, external familiarity, internal familiarity, reward / punishment, and context. These parameters not only depict the dynamic interactions amongst the trustors and trustees in an IoV network but further offer valuable insights pertinent to the behavior of the same.

3. Methods

As discussed above, the trust-based IoV dataset proposed in this particular manuscript encompasses trustors, trustees, and 9 salient trust parameters. The same are delineated as follows:

3.1. Trustor

Trust in an IoV network involves multiple attributes (parameters) which can be quantified by considering it as a relational construct involving two entities, i.e., a trustor i and a trustee j. The trustor i assumes the role of an evaluator within an IoV network to assess and ascertain the trustworthiness of a trustee j. In our proposed dataset, there are 79 trustors listed in column 1 of the dataset.

3.2. Trustee

The trustee, also referred to as a target node, is an entity that is evaluated by a trustor as either trustworthy or untrustworthy. In our proposed dataset, there are 79 trustees (listed in column 2 of the dataset) that have encountered $96,707$ interactions with the trustors.

3.3. Packet Delivery Ratio (PDR)

The Packet Delivery Ratio ($0\le PDR\le 1$) measures the degree of interaction between a trustor i and a trustee j at a time instance t within an IoV network, thereby providing a key understanding of their relationship. In order to ascertain PDR, we collect the total number of messages sent by a trustor i and successfully received by a trustee j at a time instance t in an IoV network. The PDR is, therefore, determined by taking into account the ratio between the aforementioned sent and successfully received messages between a trustor i and a trustee j. The same is listed in column 3 of the dataset, wherein 0 implies unsuccessful interaction, whereas, 1 suggests successful interaction.

3.4. Similarity (Sim)

The similarity ($0\le Sim\le 1$) between a trustor i and a trustee j at a time instance t encompasses both external similarity ($ES$) and internal similarity ($IS$), and is a weighted amalgamation of the two. The same is listed in column 4 of the dataset.

3.4.1. External Similarity (ES)

The external similarity ($0\le ES\le 1$) suggests the extent to which a trustor i and a trustee j access similar content at a time instance t, and is listed in column 5 of the dataset. ES is deemed to be 1 if the trustor i and a trustee j access similar content. Otherwise, it is regarded as 0.

3.4.2. Internal Similarity (IS)

The internal similarity ($0\le IS\le 1$) manifests the degree of similarity in the positions (geographical locations), directions (travelling trajectories), speeds, and accelerations of a trustor i and trustee j. The same is depicted in column 6 of the dataset.

3.5. Familiarity (Fam)

The familiarity ($0\le Fam\le 1$) between a trustor i and a trustee j at a time instance t is also segregated into external familiarity ($EF$) and internal familiarity ($IF$). The same is delineated in column 7 of the dataset.

3.5.1. External Familiarity (EF)

The external familiarity ($0\le EF\le 1$) quantifies the level of the familiarity a trustor possesses towards a trustee, and is listed in column 8 of the dataset. The value of $EF$ is obtained by calculating the ratio between the number of common vehicles that interact with both a trustor i and a trustee j, and the total number of vehicles that interact with a trustor over a given timestamp in an IoV network [29]. In other words, a higher number of shared interacting vehicles (i.e., $EF=1$) indicates a stronger level of familiarity between a trustor and a trustee.

3.5.2. Internal Familiarity (IF)

The internal familiarity ($0\le IF\le 1$) manifests the extent of interaction frequency between a trustor i and a trustee j, and is recorded in column 9 of the dataset. The value of $IF$ is determined by quantifying the frequency of interactions between a trustor and a trustee over a given timestamp in an IoV network. In other words, a higher interaction frequency (i.e., $IF=1$) indicates a stronger familiarity between the two parties (trustor and trustee).

3.6. Reward / Punishment (RP)

The reward / punishment ($0\le RP\le 1$) is employed in order to ascertain the degree of a reward or a penalty allocated to a trustee j based on its conduct in an IoV network. Specifically, a trustee j is rewarded by a trustor i for exhibiting cooperation, honesty, and reporting critical events, whereas, is penalized for any sort of a misconduct [30]. The RP is determined by taking into consideration the PDR, and a metric that accounts for both positive and negative interactions between a trustor and a trustee. It is thus represented in column 10 of the dataset.

3.7. Context

Context plays an indispensable role for ascertaining the trust of a trustee in an IoV network since most of the other trust parameters are directly impacted owing to the same [31]. It provides specific information regarding the settings, wherein interactions take place between a trustor i and a trustee j in an IoV network, i.e., network stability, and temporal and spatial aspects. In the context of this particular dataset, the context ($0\le Context\le 1$) implies the network communication quality segregated into four classes implying poor, medium, good, and excellent. The corresponding values pertinent to these four classes are depicted in column 11 of the dataset.

Figure 2, Figure 3, Figure 4, Figure 5 and Figure 6 depict the packet delivery ratio, similarity, familiarity, reward/punishment, and context-related scores of each of the 79 vehicles in an IoV network at their most recent respective time instance. Additionally, Table 1 delineates the values of all of the 9 trust parameters introduced in this particular dataset so as to enable the readers to have a comprehensive understanding of the same.

4. Conclusion and Future Directions

The manuscript-at-hand employs Java to design a trust-based IoV simulator which ascertains the trust values of vehicles in an IoV network and further facilitates in segregating the trustworthy vehicles from the untrustworthy ones via an optimal decision boundary. The trust-based IoV dataset obtained via this simulator is a first of its kind and encompasses 9 salient trust parameters, i.e., packet delivery ratio, similarity, external similarity, internal similarity, familiarity, external familiarity, internal familiarity, reward / punishment, and context, and provides a foundation for both the researchers from academia and industry to utilize and expand upon. In the near future, the authors intend to employ a trust-based IoV testbed to (a) ascertain the parameters introduced in this dataset via realistic interactions and (b) simulate various intricate IoV-based trust attacks, i.e., self-promoting attacks, on-off attacks, opportunistic service attacks, selective behavior attacks, bad mouthing attacks, and good mouthing attacks.