TY - JOUR
AU - Jagdale,, Balaso
AB - Abstract Routing in the Internet of Things (IoT) renders the protection against various network attacks as any attacker intrudes the routing mechanism for establishing the destructive mechanisms against the network, which insists the essentiality of the security protocols in IoT. Thus, the paper proposes a secure protocol based on an optimization algorithm, Monarch-Earthworm Algorithm (Monarch-EWA), which is the modification of the Monarch Butterfly algorithm using the Earthworm Optimization Algorithm (EWA) in order to render effective security to the network. Initially, the effective nodes are selected using the Deep Convolutional Neural Network (deep CNN) classifier based on the factors, trust and energy of the node, and stochastic gradient descent algorithm trains the deep CNN classifier. The secure nodes are involved in routing for which the secure multipath is chosen optimally using the proposed Monarch-EWA, which chooses the secure multipath based on the factors, energy and trust. The analysis of the proposed method in the presence of attacks, such as black hole, message replicate and distributed denial of service, reveals that the proposed method outperformed the existing methods. The proposed Monarch-EWA protocol acquired the maximal energy, throughput and detection rate of 0.2268 J, 48.2759% and 82.6231%, respectively, with the minimal delay of 0.0959 ms. 1. INTRODUCTION The Internet of Things (IoT) possesses several mobile or static objects, and devices are inbuilt with communication, sensors and actuators via the Internet [1, 2]. The transportation and data routing with required security and quality of services (QoS) is the major issue in IoT [2]. IoT uses various technologies, systems and evolving paradigm of devices, based on the Transmission Control Protocol/Internet Protocol, in the physical environments [3, 4]. IoT [5] finds its application in areas, such as building, agriculture and automation in management as well as industrial systems with water grids, smart grids and smart cities. The sensors in IoT are energy-constrained, performing computational and storage operations, and communicate in lossy channels. The basic driving forces in IoT represent the networking and more specifically the routing that formulates the interconnection among the IoT devices. The requirement in IoT routing [6] relies on autonomy, scalability, energy efficiency and secure communication [7, 4]. On the other hand, the unique characteristics belonging to IoT enable the nodes to be subjected to attacks. On the contrary, routing and secure communication attract as research topics in IoT [4]. Moreover, the evolution in IoT improved the implementation steps associated with the smart home networks. With easy-to-operate smart home expansion systems using IoT devices, life has become convenient, comfortable and secure [8]. IoT finds valuable application in health care, urban transportation, vehicle monitoring and space exploration, along with other potential fields [9]. Routing is a significant strategy in a communication stack, and it logically models the network in such a way that the data packets traverse along with the multiple hops between the source and destination nodes. Routing is an NP-hard problem [3], and it gains significance as the nodes in the IoT network perform as hosts and routers during data delivery to the gateways. There are many sensors in the IoT networks, and the data routing from source to destination nodes affects the power consumption of the forwarding nodes. Due to the random behavior of the network, stochastic methods are a natural fit to studying the power consumed by the individual nodes and the entire network. These methods utilize the events that occurred in the past and predicts behavior in the future. Additionally, the routing in IoT is associated with the nodes in discovering the routes with respect to the destinations through beaconing, which finally results in the overhead. At the time of beaconing, there exists flooding of route requests or ping messages of the source nodes that are rebroadcasted until the data packets reach the destination. The destination replies to the requests in order to establish the routes. Moreover, factors like beacon interval affect the transmitting rate and there is a requirement for quantization of the energy, and the performance of the protocol impacts the analysis associated with the overheads in the data packet [10, 11]. Security in IoT is a significant aspect, requiring in-depth analysis as there is a need for a secure network [12]. Trust-based methods in IOT assure secure routing, and reputation is based on the behavior of the node in the past, which is based on the degree of cooperativeness. In case of secure routing, reputation defines the forwarding and routing, and the authentication and encryption mechanisms are used which rendered the transmission of acknowledgments for a transmitted packet. Trust [12] defines the degree of confidence regarding others, and it aggregates all the reputation values held by the entity from another participant. The high reputation values of a node are marked as a trustworthy node, and the legitimate node is based on the trustworthy entity that accomplishes the communication tasks. There are systems based on trust for acquiring secure routing, and the individual system is evaluated for a specific ad hoc application and manages the constrained security threats [13]. Ensuring security in IoT is essential for a trust management mechanism (TMM), which verifies the individual requests of a service based on the security policy. TMM consists of numerous components, like secure routing, authorization and authentication. The TMM design with permitted security is a complex phenomenon [14], and the existing measure based on a rule never copes with unpredictability. The complexity associated with the interdependency in IoT insists the network rise as the key point for facilitating the security policies. The security solutions based on the network suits best for scaling the deployed IoT devices and sheer hardware diversity along with the interoperability constraints [2]. The primary intention of the research is to design and develop a secure routing protocol for IoT by proposing an optimization algorithm. Energy consumption and trust of the nodes are the two parameters considered in developing the secure routing protocol. Based on the energy consumed and the trust, a routing model, termed as Monarch-Earthworm-based secure (Monarch-EWA) routing, is introduced. The overall procedure of the proposed secure protocol includes the following four steps: in the first step, the trust level of a route is measured, and in the second step, deep learning is employed to select the secure nodes. In the third step, route discovery is performed with the route selection step in the fourth step. Initially, the energy level of the nodes is computed, followed by the computation of trust using various trust factors, like direct trust and indirect trust [15, 16], together with a new trust factor, named active trust, which is formulated newly based on a certain behavior of the nodes. Then, the deep learner is applied for selecting the secure nodes. Once these secure nodes are chosen, the route selection and route discovery are performed using the proposed Monarch-EWA algorithm. The proposed algorithm is developed by combining Monarch Butterfly Optimization (MBO) [17] with Earthworm Optimization Algorithm (EWA) [18]. The major contribution of this paper is the development of the Monarch-EWA, which is developed by integrating the MBO and EWA. The proposed Monarch-EWA selects the multipath from the total discovered paths, for secure routing. The paper is structured as follows: the background of routing protocols is discussed in Section 1, and Section 2 deliberates the review of the existing methods. The proposed method of performing secure routing is discussed in Section 3, and Section 4 gives the results of the method. Finally, Section 5 concludes the paper. 2. MOTIVATION This section deals with the existing routing protocols and the challenges of the research, which stood as the motivation for proposing a new secure protocol. 2.1. Literature survey This section reviews eight existing methods in the literature: Mick et al. [19] developed Lightweight Authentication and Secured Routing (LASeR) in which scalability was attained using a hierarchical design with minimal computational burden. The drawback of the method was about the inapplicability of the method for live testbed deployment on real devices. Hatzivasilis et al. [13] modeled a method, the Self-Channel Observation Trust and Reputation System (SCOTRES), which rendered higher degree of security, but the method failed in the presence of the jamming attackers, and no communication was accomplished, affecting the path between source and destination nodes. Shin et al. [8] modeled a lightweight and secure session key approach that assures the security of the network, and the performance was better with greater transmission rates and throughput. The drawback of the method was regarding the lack of consideration of the 5G distributed mobility management, and the performance evaluation lacked in terms of the mobility models and traffic. Airehrour et al. [4] used a time-based trust-aware RPL routing protocol (SecTrust-RPL) that acquired the optimal routing decisions even in the presence of malicious nodes, rendering an optimal throughput. The drawback of the method was that the method failed to consider secure routing since the battery of the nodes depleted and led to selfish behavior. Ai et al. [9] modeled a smart, collaborative routing protocol, Geographic energy aware routing and Inspecting Node (GIN), which minimized the network delay and reduced the packet-loss delay, thereby increasing the throughput. The failure of the method was regarding the inapplicability of the method in the information collection phenomenon. Kalkan and Zeadally [20] used a role-based security controller architecture that addressed the scalability and heterogeneity issues and rendered a network with a highly dynamic environment. However, the method suffered from security issues. Dhumane and Prasad [2] modeled a multiobjective fractional gravitational search algorithm that improves the lifetime of the network while taking into account the number of alive nodes and energy of the network. The failure of the method was about enhancing the performance by fixing the objectives of the model. Reddy and Babu [21] developed a method based on Optimal Secured Energy Aware Protocol (OSEAP) and Improved Bacterial Foraging Optimization (IBFO) algorithm which rendered a better performance with minimal energy consumption, higher throughput and delay, whereas the lifetime of the network was affected. 2.2. Challenges The challenges of the research are listed below. ▪ The significant challenge is regarding the dynamic nature of the data flow in IoT. The total users and volume of the data flow in IoT change with respect to time. However, the existing techniques for controlling the data flow assume itself as a stable network. Thus, these existing techniques did not take into account the security of the network. Whenever huge data streams reach at the same time, the whole IoT is paralyzed [22]. ▪ The significant challenges regarding the security in IoT are about their scalability and heterogeneity. In contrast to the traditional devices to assure adequate processing, computing and storage resources, IoT sensors and other mobile devices are fully resource-constrained [20]. ▪ A time-based trust-aware RPL routing protocol [4], named SecTrust-RPL, follows the mechanism based on trust. Even though the performance of the protocol was better compared with the standard RPL protocol, in case of detecting the Rank and Sybil attacks, SecTrust-RPL failed to address the colluding attacks, such as Rank/Sybil, Rank/Blackhole and Rank/Selective Forwarding attacks. ▪ Extensive devices linked to IoT suffer from challenges like large consumption of battery power, modeling of devices such as smart e-health and development of a control system for smart homes [23]. ▪ The network lifetime, failure in the battery of the IoT devices and usage of the sensor networks are the significant challenges prevailing in the existing literary works [23]. 3. PROPOSED SECURE ROUTING IN IOT BASED ON THE OPTIMAL ROUTE SELECTION Secure route selection in IoT is effective in enabling optimal communication over secure routes, which is the ultimate aim of the research. The paper discusses secure route selection based on the optimization, and Monarch-EWA chooses the multipath to progress the communication. Initially, the nodes in IoT are subjected to the calculation of trust and energy which forms the input to the secure node selection module, which is progressed using the deep CNN classifier. The input to the deep CNN classifier is the direct trust, indirect trust, active trust and energy, which are assessed for all the nodes of the network, and the secure nodes are finalized. The deep CNN classifier utilizes the SGD algorithm to tune the optimal weights, and the secure nodes are involved in the secure routing process. Using the secure nodes, the paths are discovered which are performed through fixing a source, and the destination nodes and the optimal paths are chosen based on the proposed Monarch-EWA, which is fitness-constrained. Figure 1 shows the schematic diagram of the secure routing in IoT. Consider an IoT network with |$m$| number of nodes, and let us represent the |${l}^{\mathrm{th}}$| node as |${N}_l$|⁠; |$(1\le l\le m)$|⁠. The trust and energy of the |$m$| nodes are computed, and the secure nodes are chosen based on the deep CNN, and the total secure nodes are given as |$s$|⁠, such that |$s<m$|⁠. FIGURE 1. Open in new tabDownload slide Block diagram of secure routing in IoT. FIGURE 1. Open in new tabDownload slide Block diagram of secure routing in IoT. 3.1. Deep learning networks for the secure node selection For enabling secure routing in IoT, secure nodes are to be determined, which are chosen based on the deep CNNs for which the trust and energy of the nodes are determined. The nodes with the maximal trust and maximal energy remaining in the nodes are chosen as the secure nodes by the deep CNN classifier. This section reveals the secure node selection mechanism based on the trust and energy using the deep CNN, trained with the SGD algorithm. 3.1.1. Computation of trust and energy The trust and energy computation of the nodes are summarized as follows: three types of trust—direct trust, indirect trust and active trust [15, 16]—along with the energy of the nodes are considered for formulating the secure nodes. (i) Direct trust: the direct trust of a node is calculated based on the satisfaction of a node with respect to the target. The direct trust of the |${l}^{\mathrm{th}}$| node depending on its experience is formulated as $$\begin{equation}{\mathrm{DT}}_l=\frac{1}{q}\;\sum \limits_{\begin{subarray}{l}p=1\\{}l\in p\end{subarray}}^q{W}_{lp}\end{equation}$$ (1) where |${W}_{lp}$| signifies the satisfaction degree of node |$l$| based on the |${p}^{\mathrm{th}}$| neighbors and |$q$|specifies the total neighbors of the node |$l$|⁠. It is worth interesting to note that the node |$l$|acquires the higher degree of satisfaction whenever the service provided by a node |$l$| is satisfied by node |$p$|⁠, and the satisfaction degree is based on the satisfied transaction of the neighboring node |$p$|⁠. Thus, the satisfaction degree is provided by node |$p$| to node |$l$| based on the successful transactions, which is formulated as $$\begin{equation}{W}_{lp}=\frac{T_{lp}}{m}\end{equation}$$ (2) where |${T}_{lp}$| is the success transactions of the total nodes and |$m$| specifies the total nodes in the IoT network. (ii) Indirect trust: unlike direct trust, indirect trust is based on the recommendation of the neighboring nodes, and indirect trust defines the trustworthiness of the nodes. The indirect trust of the node |$l$| depends on the recommendation of the neighboring node |$p$| along with its neighbors |$h$|⁠. Indirect trust is formulated as $$\begin{equation}{\mathrm{ID}}_l=\frac{1}{r\ast q}\;\sum \limits_{\begin{subarray}{l}p=1\\{}l\in p\end{subarray}}^r\sum \limits_{\begin{subarray}{l}h=1\\{}h\in p\end{subarray}}^q{\mathrm{sat}}_{ph}\end{equation}$$ (3) where |$r$| denotes the total neighbors of |$l$| and |$q$| specifies the neighbors of |$p$|⁠. (iii) Active trust: active trust is computed based on the data packets transmitted and received, and it is formulated as $$\begin{equation}\mathrm{AT}=\frac{1}{5}\;\left({P}_1+{P}_2+{P}_3+{P}_4\right)\end{equation}$$ (4) where |${P}_1$|⁠, |${P}_2$|⁠, |${P}_3$| and |${P}_4$| are the active trust factors determined based on the data bytes. |${P}_1$| is computed using the total bytes transmitted as $$\begin{align}{P}_1=\frac{b_v}{X}\end{align}$$ (5) where |${b}_v$| is the total bytes sent successfully and |$X$| is the total bytes sent. The second factor, |${P}_2$|⁠, is based on the received bytes, which is formulated as $$\begin{equation}{P}_2=\frac{b_u}{Y}\end{equation}$$ (6) where |${b}_u$| represents the total bytes received successfully and |$Y$| is the total bytes received by the node. The third active factor |${P}_3$| is based on the packet received with error and is formulated as $$\begin{equation}{P}_3=\left[1-\frac{\varepsilon_{rxd}}{Y}\right]\end{equation}$$ (7) where |${\varepsilon}_{rxd}$| is the package received with error and |$Y$| corresponds to the total bytes received. The fourth factor, |${P}_4$|⁠, refers to the active trust based on the error packets sent as given by $$\begin{equation}{P}_4=\left[1-\frac{\varepsilon_{\mathrm{send}}}{X}\right]\end{equation}$$ (8) where |${\varepsilon}_{\mathrm{send}}$| denotes the total number of the error bytes sent by the nodes. (iv) The energy of IoT nodes: the third parameter is the energy [2] of the IoT nodes, and energy is an essential component since the sensor nodes in IoT are operating with the energy-constrained batteries, which insist on the fact of conserving the energy of the nodes, in order to extend the network lifetime. Let us consider the initial energy of the IoT node as |${E}_0$|⁠. At the time of communication, while the transmitted data is received by the receiver, the data loss occurs, and this loss is based on the nodal characteristics and the transmission distance between the nodes in IoT. The routing protocol in the network enables the transmission, and the energy dissipation is due to the presence of the power amplifier and radio electronics in the transmitter. The energy dissipation in a node during the transmission of the data packet is modeled as in the equation below: $$\begin{equation}{E}_{\mathrm{dis}}\left({N}_l\right)={E}_{\mathrm{elc}}\times{B}_l+{E}_{\mathrm{pa}}\times{B}_l\times{\left\Vert{N}_l-{C}_g\right\Vert}^4;if\kern0.1em \left\Vert{N}_l-{C}_g\right\Vert\! \ge\! {\eta}_0\end{equation}$$ (9) where |${E}_{\mathrm{elc}}$| refers to the electronic energy, and |${E}_{\mathrm{dis}}({N}_l)$| specifies the energy dissipation of |${l}^{\mathrm{th}}$|node. The total bytes sent from the |${l}^{\mathrm{th}}$| node is represented as |${B}_l$|⁠, and |${E}_{\mathrm{pa}}$| specifies the energy corresponding to the power amplifier present in the transmitter. The energy dissipation depends on the parameter |${\eta}_0$|⁠, which is computed based on the comparison among the distance between the |${g}^{\mathrm{th}}$| head and |${l}^{\mathrm{th}}$| sensor node and the parameter |${\eta}_0$|⁠. More specifically, when the distance between the |${l}^{\mathrm{th}}$| node |${N}_l$| with respect to the CH |${C}_g$| remains below |${\eta}_0$|⁠, the energy dissipation in a normal sensor node is updated using Equation (9), or else the energy dissipation in |${N}_l$| is calculated using Equation (10). |$\Vert{N}_l-{C}_g\Vert$| notates the distance between the |${l}^{\mathrm{th}}$| node and the |${g}^{\mathrm{th}}$| cluster head. $$\begin{equation}{E}_{\mathrm{dis}}\left({N}_l\right)={E}_{\mathrm{elc}}\times{B}_l+{E}_{\mathrm{fs}}\times{B}_l\times{\left\Vert{N}_l-{C}_g\right\Vert}^2; if\kern0.1em \left\Vert{N}_l-{C}_g\right\Vert \!<{\eta}_0\end{equation}$$ (10) $$\begin{equation}{\eta}_0=\sqrt{\frac{E_{\mathrm{fs}}}{E_{\mathrm{pa}}}}\end{equation}$$ (11) where |${E}_{\mathrm{fs}}$| specifies the energy in the free space. The electrical energy depends on the modulation, coding, filtering and so on, corresponding to the transmitter and data aggregation, which is represented as $$\begin{equation}{E}_{\mathrm{elec}}={E}_{tx}+{E}_{\mathrm{agg}}\end{equation}$$ (12) where |${E}_{tx}$| refers to the transmitter energy and |${E}_{\mathrm{agg}}$| denotes the energy corresponding to the data aggregation. It is worth interesting to note that whenever a normal node |${N}_l$| in the environment attempts to communicate with |${C}_g$|⁠, energy loss occurs in the CH and this loss in energy depends on the electrical energy available at the receiver side and additionally on the data bytes received at the CH. The energy dissipation at the |${g}^{\mathrm{th}}$| CH is represented as $$\begin{equation}{E}_{\mathrm{dis}}\left({C}_g\right)={E}_{\mathrm{elec}}\times{B}_l\end{equation}$$ (13) Upon completion of the communication, the sensor nodes and CHs engaged in transmission and reception update their energy and the energy dissipated at the nodes is given as $$\begin{equation}{E}_{t+1}\left({N}_l\right)={E}_t\left({N}_l\right)-{E}_{\mathrm{dis}}\left({N}_l\right)\end{equation}$$ (14) $$\begin{equation}{E}_{t+1}\left({C}_g\right)={E}_t\left({C}_g\right)-{E}_{\mathrm{dis}}\left({C}_g\right)\end{equation}$$ (15) where |${E}_{t+1}({C}_g)$| represents the updated energy of the cluster head, |${E}_t({C}_g)$| specifies the available energy in the node at time |$t$| and |${E}_{\mathrm{dis}}({C}_g)$| is the energy dissipation in the CH or the receiver at the time of receiving the data packets from the normal node. Likewise, |${E}_{t+1}({N}_l)$| is the updated energy of the normal node, |${E}_t({N}_l)$| is the previously available energy in the normal node and |${E}_{\mathrm{dis}}({N}_l)$| symbolizes the dissipated energy in the normal nodes during transmission. The energy in the nodes and CHs continues until the node is dead with zero energy. 3.1.2. Deep convolutional neural networks for determining the secure nodes The secure nodes are determined using the SGD-based deep CNN using the energy, active trust, direct trust and indirect trust of the nodes in the environment. The necessity for enabling the secure routing in IoT insisted to evaluate the trust and energy of all the nodes and led to the selection of the nodes with maximal energy and maximal trust values. This section reveals the structure of deep CNN and the training steps. 3.1.2.1. The architecture of the deep convolutional neural network The structure of the deep CNN [24] consists of three layers, such as convolutional (conv) layers, pooling (POOL) layers and a fully connected (FC) layer, as shown in Fig. 2. The functions of the individual layers are summarized as feature extraction, sub-sampling and classification, and it is worth interesting to note that increasing the number of the conv layers enhances the detection accuracy. Generally, the features from the conv layers are fed to the fully connected layers for prediction, and in this paper, the secure nodes are determined based on the trust values and energy of the nodes. Thus, the input vector to the deep CNN is given as $$\begin{equation}\mathrm{IV}=\left\{{\mathrm{DT}}_l,{\mathrm{ID}}_l,\mathrm{AT},E\right\}\end{equation}$$ (16) where |$\mathrm{IV}$| refers to the input vector and the input factors, |${\mathrm{DT}}_l,{\mathrm{ID}}_l,\mathrm{AT},E$| correspond to the direct trust, indirect trust, active trust and energy of the nodes. FIGURE 2. Open in new tabDownload slide The architecture of deep CNN. FIGURE 2. Open in new tabDownload slide The architecture of deep CNN. Conv layers: the significant layer in deep CNN is the conv layer, which consists of a number of successive layers in such a way that the output from a layer is fed as input to the successive layers to extract the output from the fully connected layer, and at the same time, the prediction accuracy is based on the number of the conv layers. The input is fed to the conv layers, which are the energy and trust of the nodes, and from these features, more confined features are extracted, which is facilitated using the neurons in the conv layer. The neurons in the successive conv layers are linked using the set of trainable weights through the receptive fields. The convolution of the input features with the trainable weights enables the development of the feature map, which forms the input to the preceding conv layer through the nonlinear activation function. There are a set of neurons engaged in the extraction of variable features from various locations, and the neurons in the single feature map are the same, whereas the weights of the same conv layers vary. The input to the conv layers is denoted as, |$F$|⁠, and let us represent the convolution layers as $$\begin{align}G=\left\{{G}_1,{G}_2,\dots, {G}_{\mathrm{\ell}},\dots, {G}_y\right\}\end{align}$$ (17) where |$y$| corresponds to the total number of the convolutional layers. These conv layers act on the input and formulate the output, and the unit centered at |$(d,e)$| produces the output as given by $$\begin{equation}{\big({H}_{\mathrm{\ell}}^{\mathrm{\hslash}}\big)}_{d,e}={\big({Bi}_{\mathrm{\ell}}^{\mathrm{\hslash}}\big)}_{d,e}+\sum \limits_{z=1}^{f_1^{\mathrm{\ell}-1}}\sum \limits_{a=-{\omega}_1^{\mathrm{\ell}}}^{\omega_1^{\mathrm{\ell}}}\sum \limits_{c=-{\omega}_2^{\mathrm{\ell}}}^{\omega_2^{\mathrm{\ell}}}{\big({\kappa}_{\mathrm{\hslash},z}^{\mathrm{\ell}}\big)}_{a,c}\ast{\big({H}_{\mathrm{\ell}}^{\mathrm{\hslash}-1}\big)}_{d+a,e+c}\end{equation}$$ (18) where |$\ast$| specifies the convolutional operation to extract the local patterns from the output of the successive conv layers, and |${\big({H}_{\mathrm{\ell}}^{\mathrm{\hslash}}\big)}_{d,e}$| specifies the fixed feature map from conv layer |$\mathrm{\ell}$|⁠. The feature map corresponding to the previous conv layer is denoted as |${\big({H}_{\mathrm{\ell}}^{\mathrm{\hslash}-1}\big)}_{d+a,e+c}$|⁠, and |${\kappa}_{\mathrm{\hslash},z}^{\mathrm{\ell}}$|specifies the kernel function in the |${\mathrm{\ell}}^{\mathrm{th}}$| conv layers. The bias of the |${\mathrm{\ell}}^{\mathrm{th}}$| conv layer is denoted as |${Bi}_{\mathrm{\ell}}^{\mathrm{\hslash}}$|⁠, which along with the kernel function, |${\kappa}_{\mathrm{\hslash},z}^{\mathrm{\ell}}$|⁠, is determined using the SGD algorithm, and the vector to be determined is given as |$U\in \big\{ Bi,\kappa \big\}$|⁠. The activation function is used to determine the output of the neural network like yes or no. Here, the Rectified Linear Unit (ReLU) is used as the activation function to remove the negative values for simplicity and effectiveness. It is a supplementary step to the convolution operation. The computational step of a ReLU is easy, with no negative elements, no exponentials, no multiplication and no division operations. Also, ReLU is idempotent, which is very significant for deep neural networks, because each layer in the network applies nonlinearity. The |${\mathrm{\ell}}^{\mathrm{th}}$| nonlinear layer with the feature maps and the output from ReLU layer is given as $$\begin{equation}{H}_{\mathrm{\ell}}^{\mathrm{\hslash}}= fun\;\big({H}_z^{\mathrm{\hslash}-1}\big)\end{equation}$$ (19) where |$fun\;()$| is the activation function in conv layer |$\mathrm{\ell}$|⁠. Pooling (POOL) layers: convolutional networks may include local or global pooling layers to simplify the underlying computation. Pooling layers reduce the dimensions of the data by combining the outputs of neuron clusters at one layer into a single neuron in the next layer. The function of the POOL layer is fixed, and it acts as a sub-sampling layer through which the resolution of feature maps is minimized to enhance the invariance of the features to the distortions available in the input data. There is no trainable weight or bias in this layer since it is a nonparameterized layer. Fully connected layers: the abstract features from the conv layers are fed to the prediction module and the fully connected layer for the detection of secure nodes, and the output from the fully connected layer is given as $$\begin{equation}{R}_{\mathrm{\ell}}^{\mathrm{\hslash}}\!\!=\!\delta\!\;\big({V}_{\mathrm{\ell}}^{\mathrm{\hslash}}\big)\ \textrm{with}\ {V}_{\mathrm{\ell}}^{\mathrm{\hslash}}\!=\!\!\sum \limits_{z=1}^{f_1^{\mathrm{\ell}-1}}\sum \limits_{a=-{\omega}_1^{\mathrm{\ell}}}^{\omega_1^{\mathrm{\ell}}}\sum \limits_{c=-{\omega}_2^{\mathrm{\ell}}}^{\omega_2^{\mathrm{\ell}}}\!{\big({\kappa}_{\mathrm{\hslash},z}^{\mathrm{\ell}}\big)}_{a,c}\ast{\big({H}_{\mathrm{\ell}}^{\mathrm{\hslash}-1}\big)}_{d+a,e+c}\end{equation}$$ (20) where |${\kappa}_{\mathrm{\hslash},z,a,c}^{\mathrm{\ell}}$| is the weight linking the unit at |$(d,e)$| in the |${z}^{\mathrm{th}}$| feature map of layer |$(\mathrm{\ell}-1)$|and the |${\mathrm{\hslash}}^{\mathrm{th}}$| unit in layer |$\mathrm{\ell}$|⁠. Thus, the output from the deep CNN yields the secure nodes, which engage in routing to enable secure communication in IoT. Additionally, it is significant to note that deep CNN possesses a set of tunable weights and biases, which are to be determined. In this paper, the weights and biases of the deep neural network (deep CNN) is determined using the SGD algorithm [25], which is based on the minimal value of error. The training speed associated with the SGD algorithm is high with minimal cumulative error, and SGD is a stochastic approximation algorithm aiming at the minimization objective function. The advantage of SGD is that, even in the presence of large training sets, the SGD algorithms work faster as it never takes into account the entire training set; instead, it uses the required data that is responsible for computation speed and estimation. Moreover, the risk of the local minimal convergence is reduced. The following are the steps involved in the SGD algorithm. 3.1.2.2. Training phase: steps to determine the optimal weights for deep CNN The algorithmic steps for deriving the optimal weights to train the deep CNN are summarized below. (i) Initialization: the first step is initialization, and the input is represented as |$I$| with |$({\upsilon}_1,{\upsilon}_2)$| sample possessing |${\upsilon}_1$| as a |$\gamma$|-dimensional vector and |${\upsilon}_2$| which specifies the category of the training sample. (ii) Fitness evaluation: the objective function is based on the minimal function, which is given as $$\begin{equation}\operatorname{Min}\kern0.24em (U)=\frac{\nu }{2}\;{\left\Vert U\right\Vert}^2+ fun\;\left[U\;\left({\upsilon}_1,{\upsilon}_2\right)\right]\end{equation}$$ (21) The solution affording the minimal value of the fitness measure is chosen as the optimal solution for training the deep CNN classifier. (iii) Compute the weight: the weight update is based on the following equation: $$\begin{align}{U}_{\tau +1}=\left(1-\frac{1}{\tau}\right)\;{U}_{\tau }+\left({\upsilon}_2\;{\upsilon}_1\right)\end{align}$$ (22) where |${U}_{\tau +1}$| refers to the updated weight and |${U}_{\tau }$| is the weight in the previous iteration. (iv) Stopping criterion: the process of weight update is iterated until the optimal solution is derived for training the deep CNN. The output from the deep CNN is the secure nodes that are included in the routing process of the IoT networks. 3.2. Optimal multipath selection using the proposed Monarch-Earthworm algorithm for secure routing in IoT This section presents the steps for selecting the multipath using the secure nodes determined using deep CNN, and the optimal multipath selection is made through the proposed Monarch-EWA. Initially, all possible paths are determined using the Dijkstra shortest path algorithm, and the optimal multipath is decided based on the proposed Monarch-EWA using the fitness constraints, trust, energy and distance of the secure nodes. 3.2.1. Route discovery For the selection of the optimal multipath, the possible paths are formed through randomly fixing the source and destination nodes. A total of |$n$| paths are formed through the fixed source and destination nodes, and for the formation of the paths, the shortest distance between the source and destination is referred. Thus, based on the minimal distance, the paths are formed, which are selected using the fitness constraints based on Monarch-EWA. 3.2.2. Solution encoding of the Monarch-Earthworm algorithm The solution vector is the vector with |$k$| paths, which varies between 1 and |$n$|⁠, where |$n$| is the total number of paths generated using the Dijkstra shortest path algorithm. Each path comprises of a source and destination nodes along with intermediate nodes involved in the secure routing phenomenon of the IoT network. Figure 3 shows the solution vector with |$k$| paths. As an instance, consider a total of |$k$| paths with a total of |$n$| nodes that take the value 10. Each path in the solution vector comprises a source node |$S$|⁠, destination node |$D$| and intermediate nodes, and in Path 1, there are three intermediate nodes, Node 3, Node 5 and Node 9. FIGURE 3. Open in new tabDownload slide Solution vector. FIGURE 3. Open in new tabDownload slide Solution vector. 3.2.3. Fitness measure The fitness of the solution is computed based on three constraints, energy, distance and trust. The fitness measure is given as $$\begin{equation}F=\frac{1}{2}\;\left[\sum \limits_{x=1}^k\sum \limits_{l=1}^{\left|s\right|}\left({Tt}_{x,l}+{E}_{x,l}+\left(1-{Dt}_{x,l}\right)\right)\right]\end{equation}$$ (23) where |${Tt}_{x,l}$| is the trust of the |${l}^{\mathrm{th}}$| secure node in the |${xx}^{\mathrm{th}}$| path, |${E}_{x,l}$| specifies the energy of the |${l}^{\mathrm{th}}$| secure node in the |${x}^{\mathrm{th}}$| path and |${Dt}_{x,l}$| denotes the distance of the |${l}^{\mathrm{th}}$| secure node in the |${x}^{\mathrm{th}}$| path. There are a total of |$k$| paths and |$|s|$| secure nodes. The trust and the energy remaining in the nodes should be maximal, whereas the distance measure should be minimal. The trust of the node in the |${x}^{\mathrm{th}}$| path is computed as $$\begin{equation}{Tt}_{x,l}=\frac{1}{3}\;\left[{\mathrm{DT}}_{x,l}+{\mathrm{ID}}_{x,l}+{\mathrm{AT}}_{x,l}\right]\end{equation}$$ (24) The distance measure should be minimal, and since the fitness is aimed at maximum, the distance measure |${Dt}_{x,l}$| is converted as a maximization problem by subtracting |${Dt}_{x,l}$| from 1. Thus, the distance is computed as $$\begin{equation}{Dt}_{xx,l}=\frac{\mathrm{Dist}\;\left(x,l\right)}{Nf}\end{equation}$$ (25) where |$Nf$| corresponds to the normalization factor and |$\mathrm{Dist}\;(x,l)$| refers to the node in the |${x}^{\mathrm{th}}$| path. The fitness is a maximization problem, and the solution affording the maximal value of the fitness is chosen as the best solution or best multipath to progress the secure routing. 3.2.4. Monarch-EWA for selecting the secure routing The multipath selection is made through Monarch-EWA, which is the integration of the standard EWA [18] in the standard MBO [17]. The standard MBO is based on the migration behavior of the monarch butterflies in which there are two populations of butterflies, one living in Land 1 and other living in Land 2. There are two optimization parameters, migration operator and adjusting operator. The prime goal to choose the MBO algorithm is that it is simple and does not involve any complex computations, and the effort required for fine-tuning the optimization parameters is better. Additionally, the search speed is high, and the performance is better. However, in the case of low-dimensional functions, the performance is poor, which is enhanced through the integration of EWA in MBO. Additionally, EWA is more effective for parallel computation and enables the effective balance between the exploration and exploitation phases. Moreover, converging to the local optimal is reduced, implementing the global optimal feature in MBO. Hence, the proposed Monarch-EWA has the advantages of both MBO and EWA, selecting the optimal paths for secure routing. The progressing steps of Monarch-EWA are as follows. Step 1: Initialization: the solution is initialized as |${K}_{i,j}^{\tau }$| with |$Q$| being the total population in search space. The generation counter is fixed as |$gc=1$|⁠, and the solutions are randomly chosen with |${Q}_1$| and |${Q}_2$| being the two monarch butterfly populations in Land 1 and Land 2, respectively. Let us denote the maximal step size |$Ms$|⁠, adjusting the rate |$Ar$|⁠, migrating rate |$Mt$| and migration ratio |$Mr$|⁠. Step 2: Compute the fitness of solutions: the best solution is chosen based on the maximal value of the fitness as is shown in Equation (23), and the equations are sorted according to the maximal value of the fitness. Step 3: Update the position of monarch butterflies: the solutions are updated based on the migration and adjusting operators, for which the populations are sub-divided as |${Q}_1$| and |${Q}_2$|⁠. For the sub-population |${Q}_1$|⁠, the migration operator is used for position update, and the adjusting operator is employed for position update in sub-population |${Q}_2$|⁠. In the case of the migration operator, whenever the position of the monarch butterfly lies below the ratio of monarch butterflies in Land 1, then the position of a butterfly is updated. Additionally, the butterfly updates the position based on the adjusting operator as shown below: $$\begin{equation}{K}_{i,j}^{\tau +1}={K}_{i,j}^{\tau }+\beta \ast \left({w}_j-0.5\right)\end{equation}$$ (26) where |$\beta$| is the weighting factor, |${w}_j$| denotes the walk step of the monarch and |${K}_{i,j}^{\tau }$| specifies the position of the |${j}^{\mathrm{th}}$| monarch butterfly and |$i$| refers to the |${i}^{\mathrm{th}}$| element in the |${j}^{\mathrm{th}}$|monarch butterfly. The above equation for the adjusting operator is modified using EWA to inherit the advantages of EWA in Monarch-EWA. The standard equation of EWA is given as $$\begin{equation}{K}_{i,j}^{\tau +1}={K}_{i,j}^{\tau }+{x}_j\ast M\end{equation}$$ (27) where |$M$| is the random number and |${x}_j$| refers to the weight vector. For deriving the proposed Monarch-EWA, Equation (26) is multiplied with 2 as $$\begin{equation}2\times{K}_{i,j}^{\tau +1}=2\times{K}_{i,j}^{\tau }+2\times \beta \ast \left({w}_j-0.5\right)\end{equation}$$ (28) $$\begin{equation}{K}_{i,j}^{\tau +1}+{K}_{i,j}^{\tau +1}=2\times{K}_{i,j}^{\tau }+2\times \beta \ast \left({w}_j-0.5\right)\end{equation}$$ (29) Substituting Equation (27) with (29), we get $$\begin{equation}{K}_{i,j}^{\tau +1}+{K}_{i,j}^{\tau }+{x}_j\ast M=2\times{K}_{i,j}^{\tau }+2\times \beta \ast \left({w}_j-0.5\right)\end{equation}$$ (30) $$\begin{equation}{K}_{i,j}^{\tau +1}={K}_{i,j}^{\tau }+2\times \beta \ast \left({w}_j-0.5\right)-{x}_j\ast M\end{equation}$$ (31) Thus, Equation (31) is the updated equation of the proposed Monarch-EWA such that the solution is updated based on the best position in the previous iteration and on the weighting factor. Step 4: Termination: the steps are repeated for the maximal number of the iterations until the global optimal solution (multipath) is derived. Algorithm 1 shows the pseudo code of Monarch-EWA. Open in new tabDownload slide Open in new tabDownload slide 4. RESULTS AND DISCUSSION This section presents the results of the proposed Monarch-EWA protocol and the comparative analysis of the methods to reveal the effectiveness of the protocol. 4.1. Experimental setup The simulation of IoT is done using MATLAB with ThingSpeak with different parameters like simulation area, number of nodes, energy, mobility model and propagation model. Table 1 shows the simulation parameters of the IoT network. 4.2. Experimental results The sample results of the experiment are depicted in this section, and Fig. 4 shows the sample results of the experiment. The sample simulation results at Round_10, Round_20, Round_25 and Round_30 are depicted in Fig. 4a, Fig. 4b, Fig. 4c and Fig. 4d, respectively. The packet transmission between the source and destination nodes is depicted in Fig. 4; the red color round indicates the source node and the green color round symbolizes the destination, and at the same time, the full round is the nodes in the simulation environment. TABLE 1. Simulation parameters. Variables Values Maximum communication radio range 100 × 100 m2 Initial energy 0.5 J Receiver energy 5 × 10−7 Energy of multipath fading model 13 × 10−15 pJ/bit/m2 Transmitter energy 10 × 10−7 nJ/bit/m2 Energy of free space 1 × 10−10 pJ/bit/m2 Data aggregation energy 5 × 10−8 nJ/bit/signal Mobility span 100 m Speed 0.0025 m/round Variables Values Maximum communication radio range 100 × 100 m2 Initial energy 0.5 J Receiver energy 5 × 10−7 Energy of multipath fading model 13 × 10−15 pJ/bit/m2 Transmitter energy 10 × 10−7 nJ/bit/m2 Energy of free space 1 × 10−10 pJ/bit/m2 Data aggregation energy 5 × 10−8 nJ/bit/signal Mobility span 100 m Speed 0.0025 m/round Open in new tab TABLE 1. Simulation parameters. Variables Values Maximum communication radio range 100 × 100 m2 Initial energy 0.5 J Receiver energy 5 × 10−7 Energy of multipath fading model 13 × 10−15 pJ/bit/m2 Transmitter energy 10 × 10−7 nJ/bit/m2 Energy of free space 1 × 10−10 pJ/bit/m2 Data aggregation energy 5 × 10−8 nJ/bit/signal Mobility span 100 m Speed 0.0025 m/round Variables Values Maximum communication radio range 100 × 100 m2 Initial energy 0.5 J Receiver energy 5 × 10−7 Energy of multipath fading model 13 × 10−15 pJ/bit/m2 Transmitter energy 10 × 10−7 nJ/bit/m2 Energy of free space 1 × 10−10 pJ/bit/m2 Data aggregation energy 5 × 10−8 nJ/bit/signal Mobility span 100 m Speed 0.0025 m/round Open in new tab FIGURE 4. Open in new tabDownload slide Sample results of the experiment: (a) Round_10, (b) Round_20, (c) Round_25 and (d) Round_30 FIGURE 4. Open in new tabDownload slide Sample results of the experiment: (a) Round_10, (b) Round_20, (c) Round_25 and (d) Round_30 4.3. Performance metrics The performance metrics used for the comparison include the detection rate, throughput, delay and energy. The throughput of the network refers to the total data rates transmitted over the network per second, and delay refers to the time taken for the transmission of the data irrespective of the attacker. The metric energy refers to the energy that remained in the nodes at the end of the iteration, the energy that remained in the nodes is maximum for the effective method and the detection rate refers to the accurate detection of the attacker. The effective method reports with the maximal detection rate, maximal energy, maximal throughput and minimal delay. 4.4. Competing methods The methods used for comparison includes MBO [17], LASeR [19] and SCOTRES [13], which is compared with the proposed Monarch-EWA to prove the effectiveness of the proposed method of secure routing. 4.5. Comparative analysis In this section, a comparative analysis of the methods is presented, and the analysis is performed based on the performance metrics in the presence of three attacks. 4.5.1. Comparative analysis in the presence of a black hole attack Figure 5 shows the analysis of the methods in the presence of a black hole attack. Figure 5a shows the analysis based on the delay (ms) with respect to the number of rounds. When the number of the round is 200, the delay of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 0.1008, 0.1176, 0.1323 and 0.0976 ms, respectively. Figure 5b shows the analysis based on the detection rate (%) with respect to the number of rounds. When the number of the round is 200, the detection rate of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 86.2690, 87.4296, 86.8493 and 92.3620%, respectively. Figure 5c shows the analysis based on the energy (J) with respect to the number of rounds. When the number of the round is 200, the energy of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 0.4470, 0.4429, 0.4271 and 0.4557 J, respectively. Figure 5d shows the analysis based on the throughput (%) with respect to the number of rounds. When the number of the round is 200, the throughput of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 92.742, 99.227, 93.739 and 100%, respectively. FIGURE 5. Open in new tabDownload slide Comparative analysis in the presence of a black hole attack: (a) delay, (b) detection rate, (c) energy and (d) throughput FIGURE 5. Open in new tabDownload slide Comparative analysis in the presence of a black hole attack: (a) delay, (b) detection rate, (c) energy and (d) throughput 4.5.2. Comparative analysis in the presence of a DDOS attack Figure 6 shows the analysis of the methods in the presence of a distributed denial-of-service (DDOS) attack. Figure 6a shows the analysis based on the delay (ms) with respect to the number of rounds. When the number of the round is 200, the delay of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 0.2016, 0.0840, 0.0976 and 0.0735 ms, respectively. Figure 6b shows the analysis based on the detection rate (%) with respect to the number of rounds. When the number of the round is 200, the detection rate of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 86.301, 86.881, 88.042 and 92.105%, respectively. Figure 6c shows the analysis based on the energy (J) with respect to the number of rounds. When the number of the round is 200, the energy of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 0.4649, 0.4612, 0.4444 and 0.4660 J, respectively. Figure 6d shows the analysis based on the throughput (%) with respect to the number of rounds. When the number of the round is 200, the throughput of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 92.8947, 98.9696, 93.41176 and 100%, respectively. FIGURE 6. Open in new tabDownload slide Comparative analysis in the presence of a DDOS attack: (a) delay, (b) detection rate, (c) energy and (d) throughput FIGURE 6. Open in new tabDownload slide Comparative analysis in the presence of a DDOS attack: (a) delay, (b) detection rate, (c) energy and (d) throughput 4.5.3. Comparative analysis in the presence of a message replicate attack Figure 7 shows the analysis of the methods in the presence of a message replicate attack. Figure 7a shows the analysis based on the delay (ms) with respect to the number of rounds. When the number of the round is 200, the delay of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 0.1209, 0.0840, 0.1367 and 0.0735 ms, respectively. Figure 7b shows the analysis based on the detection rate (%) with respect to the number of rounds. When the number of the round is 200, the detection rate of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 90.702, 89.506, 90.104 and 94.290%, respectively. Figure 7c shows the analysis based on the energy (J) with respect to the number of rounds. When the number of the round is 200, the energy of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 0.439, 0.431, 0.416 and 0.441 J, respectively. Figure 7d shows the analysis based on the throughput (%) with respect to the number of rounds. When the number of the round is 200, the throughput of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is 93.333, 100, 93 and 100%, respectively. FIGURE 7. Open in new tabDownload slide Comparative analysis in the presence of message replicate attack: (a) delay, (b) detection rate, (c) energy and (d) throughput FIGURE 7. Open in new tabDownload slide Comparative analysis in the presence of message replicate attack: (a) delay, (b) detection rate, (c) energy and (d) throughput 4.6. Comparative discussion This section presents a comparative discussion of the methods in the presence of the attacks, and the values for the delay, energy, throughput and detection rate are visualized at the end of the iteration. The delay of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is found to be minimal when a DDOS attack is presented in the IoT network, and the delay is 0.2630, 0.1096, 0.1274 and 0.0959 ms for the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA, respectively. It is clear from Table 2 that the delay of the proposed method is minimal compared with the existing methods. The detection rate of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is found to be maximal when the message replicate attacks is present in the IoT network and the detection rate is 81.718%, 80.6404%, 81.1792%, and 82.6231% for the methods, MBO, LASeR, SCOTRES, and proposed Monarch-EWA, respectively. It is clear from the Table 2 that the proposed method acquired a maximal detection rate compared with the existing methods. The energy of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is found to be maximal when a DDOS attack is presented in the IoT network, and the energy is 0.1975, 0.1800, 0.1055 and 0.2268 J for the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA, respectively. It is clear from Table 2 that the proposed method possessed a maximal energy compared with the existing methods at the end of 1000 rounds. The throughput of the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA is found to be maximal when a message replicate attack is presented in the IoT network, and the throughput is 31.0638, 37.1739, 36.0000 and 48.2759% for the methods, MBO, LASeR, SCOTRES and proposed Monarch-EWA, respectively. It is clear from Table 2 that the proposed method possessed a maximal throughput compared with the existing methods at the end of 1000 rounds. From the table, it can be exposed that the proposed Monarch-EWA has the best performance by obtaining minimum delay, maximum detection rate, energy and throughput. The reasons for the performance improvement of the proposed method are as follows: the proposed method considers the energy and trust factors for choosing the routing path. Although several factors have been considered for secure routing, trust and energy are the most important factors. The trust can be considered as a significant factor because the survival of an IoT is dependent upon the cooperative and trusting nature of its nodes. Energy awareness enables load balancing, which is important for both the elongation of the network lifetime and the defense against traffic analysis attacks. Hence, the proposed method offers a more secure path for data transmission. Also, the proposed method has the advantages of both MBO and EWA. MBO is simple and does not involve any complex computations, and EWA is more effective for parallel computation. TABLE 2. Discussion table. Metrics (for Round_1000) MBO LASeR SCOTRES Proposed Monarch-EWA Delay (ms) 0.2630 0.1096 0.1274 0.0959 Detection rate (%) 81.718 80.6404 81.1792 82.6231 Energy (J) 0.1975 0.1800 0.1055 0.2268 Throughput (%) 31.0638 37.1739 36.0000 48.2759 Metrics (for Round_1000) MBO LASeR SCOTRES Proposed Monarch-EWA Delay (ms) 0.2630 0.1096 0.1274 0.0959 Detection rate (%) 81.718 80.6404 81.1792 82.6231 Energy (J) 0.1975 0.1800 0.1055 0.2268 Throughput (%) 31.0638 37.1739 36.0000 48.2759 Open in new tab TABLE 2. Discussion table. Metrics (for Round_1000) MBO LASeR SCOTRES Proposed Monarch-EWA Delay (ms) 0.2630 0.1096 0.1274 0.0959 Detection rate (%) 81.718 80.6404 81.1792 82.6231 Energy (J) 0.1975 0.1800 0.1055 0.2268 Throughput (%) 31.0638 37.1739 36.0000 48.2759 Metrics (for Round_1000) MBO LASeR SCOTRES Proposed Monarch-EWA Delay (ms) 0.2630 0.1096 0.1274 0.0959 Detection rate (%) 81.718 80.6404 81.1792 82.6231 Energy (J) 0.1975 0.1800 0.1055 0.2268 Throughput (%) 31.0638 37.1739 36.0000 48.2759 Open in new tab 5. CONCLUSION The secure routing in IoT is performed using the secure routing protocol, Monarch-EWA, which is the integration of the EWA in MBO in such a way that the throughput and energy of the nodes are not affected even in the presence of the attacks. The attacker detection rate is found to be better for the proposed method with a better performance. To enable the security in the network, the trust and energy of the nodes are computed, and the secure nodes are chosen using the SGD-based deep CNN. The secure nodes are involved in the multipath routing for which the optimal multipath is chosen based on the proposed Monarch-EWA using the constraints, such as energy, distance and trust of the nodes in the path, and it is significantly important to use any path generation algorithm for which the Dijkstra shortest path algorithm is used. Thus, the paths generated using the Dijkstra shortest path algorithm is optimally chosen using the proposed algorithm for enabling the secure routing in the network. The simulation of the IoT network reveals that the proposed method outperformed the existing methods with maximal energy, throughput and detection rate of 0.2268 J, 48.2759%, and 82.6231%, respectively, and a minimal delay of 0.0959 ms. S.S. received a degree in Computer Science & Engineering from P.E.S College of Engineering affiliated to Dr B.A.M. University Pune. He received MTech in Network & Internet Engineering from S.J.C.E Mysore, V.T. University, in 2005. He is awarded with a Ph.D. in the field of Computer Science & Engineering from J.N.T.U Anantapur, India. He is presently working as an Assistant Professor at the Department of CSE at MIT World Peace University, Pune, India. He has more than 13 years of teaching and industry experience. He has publications in journals and conference proceedings. His research interest is Software Engineering & IOT data life cycle including its security. B.N.J. received a diploma in Industrial Electronics from Govt. Polytechnic, Latur, and BE Computer Engineering degree from Pune University in 1992. He received ME in Computer Engineering, from VJTI, under Mumbai University in 1999. He is awarded with a Ph.D. in the field of Information Security from GH Raisoni College of Engineering affiliated to RTM Nagpur University, India. He is presently working as an Associate Professor at the Department of CSE at MIT World Peace University, Pune, India. He has more than 27 years of academics experience including head of the computer department at Sardar Patel College of Engineering, Mumbai. He has publications in journals, conference proceedings and books. His research interests are IOT data life cycle and its security. He has also a Certified Ethical Hacker certification from the EC Council in his credit. He is a member in Govt. committees, University faculty interview- Subject Expert in Pune and Mumbai University. References [1]. Bouaziz , M. and Rachedi , A. ( 2016 ) A survey on mobility management protocols in wireless sensor networks based on 6LoWPAN technology . Comput. Commun. , 74 , 3 – 15 . Google Scholar Crossref Search ADS WorldCat [2]. Dhumane , A.V. and Prasad , R.S. ( 2017 ) Multi-objective fractional gravitational search algorithm for energy efficient routing in IoT . Wirel. Netw. , 1 – 15 . WorldCat [3]. Watteyne , T. , Molinaro , A. , Richichi , M.G. and Dohler , M. ( 2010 ) From MANET to IETF ROLL standardization: a paradigm shift in WSN routing protocols . IEEE Commun. Surv. Tut. , 13 , 688 – 707 . Google Scholar Crossref Search ADS WorldCat [4]. Airehrour , D. , Gutierrez , J.A. and Ray , S.K. ( 2018 ) SecTrust-RPL: a secure trust-aware RPL routing protocol for Internet of Things . Future Gener. Comput. Syst. WorldCat [5]. shende , D. ( 2018 ) Automation of dry-wet waste collection to support Swachh Bharat Abhiyan and its monitoring over IOT enabled WSN . Int. J. Comput. Sci. En. , 6 , 477 – 479 . WorldCat [6]. Shende , D.K. and Nikhil , S. ( 2018 ) IoT based geographic multicast routing protocol with DPA through WSN . International Journal of Creative Research Thoughts , 6 , 578 – 584 . WorldCat [7]. Huia , T.K.L. , Sherratt , R.S. and Sánchez , D.D. ( 2017 ) Major requirements for building Smart Homes in Smart Cities based on Internet of Things technologies . Future Gener. Comput. Sys. , 76 , 358 – 369 . Google Scholar Crossref Search ADS WorldCat [8]. Shin , D. , Sharma , V. , Kim , J. , Kwon , S. and You , I. ( 2017 ) Secure and efficient protocol for route optimization in PMIPv6-based Smart Home IoT networks . IEEE Access , 5 , 11100 – 11117 . Google Scholar Crossref Search ADS WorldCat [9]. Ai , Z. , Zhou , Y. and Song , F. ( 2018 ) A smart collaborative routing protocol for reliable data diffusion in IoT scenarios . 18 . [10]. Sankaran , S. and Sridhar , R. ( 2015 ) Modeling and Analysis of Routing in IoT Networks. In Proceedings of International Conference on Computing and Network Communications (CoCoNet) . [11]. Dhumane , A. , Prasad , R. and Prasad , J. ( 2016 ) Routing issues in internet of things: a survey . Proceedings of the International Multiconference of Engineers and Computer Scientists , 1 , 16 – 18 . [12]. Hatzivasilis , G. and Manifavas , C. ( 2012 ) Building Trust in Ad Hoc Distributed Resource-Sharing Networks Using Reputation-Based Systems . In Proceedings of 16th Panhellenic Conference on Informatics . [13]. Hatzivasilis , G. , Papaefstathiou , I. and Manifavas , C. ( 2017 ) SCOTRES: secure routing for IoT and CPS . IEEE Internet Things , 4 , 2129 – 2141 . Google Scholar Crossref Search ADS WorldCat [14]. Hashemi , S.M. and He , J. ( 2017 ) LA-based approach for IoT security, journal of robotics . Netw. Artif. Life , 3 , 240 – 248 . Google Scholar Crossref Search ADS WorldCat [15]. Das , A. and Islam , M.M. ( 2012 ) SecuredTrust: a dynamic trust computation model for secured communication in multi-agent systems . IEEE T. Depend. Secure Comput. , 9 . WorldCat [16]. Wang , B. , Chen , X. and Chang , W. ( 2014 ) A light-weight trust-based QoS routing algorithm for ad hoc networks . Pervasive Mob. Comput. , 13 , 164 – 180 . Google Scholar Crossref Search ADS WorldCat [17]. Wang , G. , Deb , S. and Cui , Z. ( 2015 ) Monarch butterfly optimization . Neural Comput. Appl. , 1 – 20 . WorldCat [18]. Wang , G. , Coelho , L. and Deb , S. ( 2015 ) Earthworm optimization algorithm: a bio-inspired metaheuristic algorithm for global optimization problems . Int. J. Bio-Inspir. Com. . WorldCat [19]. Mick , T. , Tourani , R. and Misra , S. ( 2018 ) LASeR: Lightweight Authentication and Secured Routing for NDN IoT in Smart Cities . IEEE Internet Things J. , 5 , 755 – 764 . Google Scholar Crossref Search ADS WorldCat [20]. Kalkan , K. and Zeadally , S. ( 2017 ) Securing Internet of Things (IoT) with Software Defined Networking (SDN) . IEEE Commun. Mag. , 1 – 7 . WorldCat [21]. Reddy , P.K. and Babu , R. ( 2017 ) An evolutionary secure energy efficient routing protocol in Internet of Things . Int. J. Intell. En. Syst. , 10 , 337 – 346 . WorldCat [22]. Liu , Y. , Kuang , Y. , Xiao , Y. and Xu , G. ( 2018 ) SDN-based data transfer security for Internet of Things . IEEE Internet Things J. , 5 , 257 – 268 . Google Scholar Crossref Search ADS WorldCat [23]. Shah , S.B. , Chen , Z. , Yin , F. , Khan , I.U. and Ahmad , N. ( 2018 ) Energy and interoperable aware routing for throughput optimization in clustered IoT-wireless sensor networks . Future Gener. Comput. Syst. , 81 , 372 – 381 . Google Scholar Crossref Search ADS WorldCat [24]. Tu , F. , Yin , S. , Ouyang , P. , Tang , S. , Liu , L. and Wei , S. ( 2017 ) Deep convolutional neural network architecture with reconfigurable computation patterns . IEEE Transactions on Very Large Scale Integration (VLSI) Systems , 25 . WorldCat [25]. Yang , J. and Yang , G. ( 2018 ) Modified convolutional neural network based on dropout and the stochastic gradient descent optimizer , 11 ( 3 ). © The British Computer Society 2019. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Monarch-EWA: Monarch-Earthworm-Based Secure Routing Protocol in IoT
JF - The Computer Journal
DO - 10.1093/comjnl/bxz135
DA - 2020-06-18
UR - https://www.deepdyve.com/lp/oxford-university-press/monarch-ewa-monarch-earthworm-based-secure-routing-protocol-in-iot-7XjPL0H8FQ
SP - 1
VL - Advance Article
IS - 
DP - DeepDyve
ER -