Implementing adaptive dragonfly optimization for privacy preservation in IoT

Abstract

IoT (Internet of Things) is a sophisticated analytics and automation system that utilizes networks, big data, artificial intelligence, and sensing technology, and is controlled by an embedded module. It allows one to use affordable wireless technology and transmits the data into the cloud at a component level. It also provides a place to save the data – however, the significant challenges in IoT relay on security restrictions related with device cost. Moreover, the increasing amount of devices further generate opportunities for attacks. Hence, to overcome this issue, this paper intends to develop a promising methodology associated with data privacy preservation for handling the IoT network. It is obvious that the IoT devices often generate time series data, where the range of respective time series data can be vast. Under such circumstances, proper information extraction through effective analysis and relevant privacy preservation of sensitive data from IoT is challenging. In this paper, the problem that occurred in the data preservation is formulated as a non-linear objective model. To solve this objective model, an improved, optimized Dragonfly Algorithm (DA) is adopted, which is termed the Improved DA (IDA) algorithm. Here, the proposed model focused on preserving the physical activity of human monitoring data in the IoT sector. Moreover, the proposed IDA algorithm is compared with conventional schemes such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Bee Colony (ABC), Firefly (FF) and DA and the outcomes prove that the suggested scheme is highly used for preserving the sensitive information uploaded in IoT.

Keywords

Internet of Things sensitive data privacy preservation Improved Dragonfly Algorithm

1. Introduction

The IoT is a compilation of physical objects that are entrenched with software, electronics, together with sensors, which allows objects to be sensed and controlled distantly across the traditional network infrastructure, makes the direct integration feasible among computer communication networks [1,17,26,29]. Therefore, IoT is generally applied in numerous applications such as environment monitoring, privacy information [28] transportation, medical healthcare systems [7,10,11,37], building automation, and energy management [21,31]. Moreover, IoT is the contemporary web evolution that comprises billions of devices, which are preserved by various association and people, who are employed to utilize them for their own determinations [3–5]. It can hold the data of entire countries [9]. Further, cyber security and privacy manages with the embarrassment of intimidations which currently interrupts the organizations. IoT must deal powerfully with such intimidations and confidentiality of the information gathered and assure the protection and are distilled from IoT strategies to comprehend its entire potential [2,27]. On the other hand, IoT offers several characteristic limitations that make the appliance of traditional privacy methods and security challenges [39]. It is owing to the IoT solutions, which comprise a variety of private security and solutions for defending such IoT data in addition to the store at the layer of the device, the IoT platform and the infrastructure layer or IoT application layer [42]. Subsequently, a magnificent confront in IoT is to guarantee the end-to-end security across the mentioned three IoT layers.

Unsuitably, owing to the resource constrictions of IoT devices, it hands over tremendously multifaceted computation to the energy abundant cloud for significantly improved capability forever [35]. On the other hand, the outputs, inputs, in addition to the role of the fundamental estimation might be intimately related to the privacy of IoT users, which could not be undefended to collusion between malicious IoT users in addition to malicious cloud servers [13,14]. Several cases take place, where information and communication systems are endangered by one who utilizes security guidelines in Information and Communication Technology (ICT) appliances [12,22]. Therefore, several security tools are being developed and researched to defend from communication and information attacks. The protection requirements and susceptibility for protocols that are utilized in an atmosphere are scrutinized [40]. In addition to security solutions, the related threat investigation is suggested in [19,25] with the intention of dealing with security threats using network edge strategies. In [23,32] an examination of an IoT environment is provided with major security associated factors. As a result, an important problem of immense concern is necessary for how to model new proficient privacy-preserving solutions with IoT–cloud convergence for subsequent generation mobile technologies.

This paper contributes an enhanced data sanitization and data restoration over the IoT network for secure data transmission. Primarily, a key is produced by hiding the secured data, by which the sensitive data is sanitized in a protected way, and the data restoration is performed using the inverse key. For an improved key generation, an optimization scheme known as DA is employed to obtain the objective model. Besides, the suggested scheme is distinguished with various traditional algorithms like GA, PSO, ABC, FF and DA algorithms and the outcomes are attained. This approach is feasible to preserve all types of datasets in different applications like medical, business, entertainment, etc.

The paper is organized as follows. Section 2 explains the related works and reviews done on this topic. Section 3 discusses the proposed data privacy preservation architecture and Section 4 analyses the optimal key generation based on the DA algorithm. Section 5.1 describes the simulation procedure, and Section 5 portrays the results and discussions. Finally, Section 6 concludes the paper.

2. Literature review

2.1. Related works

In 2017, Prem Prakash Jayaraman et al. [16] proposed IoT for end-to-end cyber security to prevail over the threats in privacy. This has been attained by initiating novel methods to conserve the IoT data, obtaining IoT architectures and performance of concept systems, which guarantees the confidentiality of IoT data that, was widened for open source platform. The entire data was gathered arbitrarily from the IoT tool and articulated as data addends when accumulating one of the numbers in the component. The effectiveness and presentation were estimated, and applicability and possibility on the open platform were confirmed experimentally. The suggested method had improved performance on evaluating the proposed IoT system since it produces the data and conserves the system privacy.

In 2017, Gang Sun et al. [30] hypothetically evaluated the Attack Dummy Local Selection (ADLS) scheme with their suggested Dummy Local Selection (DLS) algorithm. This was made on the location Based Services (LBS), in which the user privacy was attacked. LBS offer service to the entire users offering the improved probability to follow the data of third parties. Using the offered information and the metric entropy, DLS chooses a dummy location for best privacy level. ADLS scheme has the benefit of recognizing the location of the user from the dummy DLS location. On executing the algorithms, they establish that the implemented algorithm offers data-driven service of IoT with improved effectiveness and reduced performance error defending the user’s location.

In 2017, Mujahid Mohsin et al. [24] formed a novel and a comprehensive data-driven service IoT checker that was scalable; interoperable offering enhanced security. The variances of security formation were arrested routinely, and the intimidation vectors were evaluated. Real world IoT commodities were calculated for security depending on ontology. The human formation errors were moreover eradicated. Automatic security arrangement for a variety of global specifications, interactions, and dependencies was offered. Threats vectors develop from several cyber privacy attacks were also resolved. Simulation of rule dependent constraints using several configurations exposed the performance of the network such as scalability and effectiveness the size of the network.

In 2017, Seokcheol Lee et al. [20] inspected the susceptibilities in the existing security system. A Game Theory dependent on IoT was suggested to prevail over the issues like adverse effects owing to the mutual communication of adversary, loss of objectivity in cost efficiency, and security administrator. It encompasses three steps specifically analyzing cost, strategy modeling, and payoff computation. To construct a safe IoT, a network location is considered, therefore the crisis and the solution can be explained. The network was safeguarded against a variety of cyber-attacks. The proposed technique was evaluated with the legacy scheme. The security susceptibilities were noticed, and moreover, it eradicates the inconveniences of cyber-attack.

In 2017, Ming Tao et al. [33] worked on the highly developed Cloud computing technology and IoT for offering enhanced service in home mechanization. They measured interoperations, of interaction heterogeneous devices that produced a cloud layered platform which is dependent on ontology for offering security service framework. This scheme has facilitated monitoring and remote controlling of business intelligence, home applications, and multimedia entertainment for a smart home ecological unit. The intention of this technique was to offer pleasant and secured smart monitoring and control of the home appliances. It has moreover offered the cybersecurity for all the communications. Certain inconveniences such as deficient in globalizing the standards owing to the multifaceted addressing were produced from collaboration of the home devices with any government authorities or organization.

In 2017, Bogdan-Cosmin Chifor et al. [6] presented a scheme that offers a secured communication channel among the cloud computing services and the embedded system. It is a user-centered tool giving security from a cloud system, which was not confidential. A security stack was incorporated in conventional IoT frame previously for heterogeneous devices, which was offered to the user to substantiate their strategies. Linkage among various accounts could not be able to be done by the producer. A device authorization system attaches the smart home IoT applications to a cloud, which was not confidential. The FIDO protocol was adopted for transferring secured authoritative messages and unlocking was made by the user employing a private key, such as biometric authentication.

In 2017, Mengmeng Ge et al. [8] designed a structure to protect the IoT that includes five phases namely, model updates, security visualization, security model generation, data processing and security analysis. The probable improvement and the consequence of the security attack were learned. The structure was produced from IoT generator that offers information network. Moreover, security modeled generator offering the comprehensive Security Evaluator and Hierarchal Attack Representation Model (HARM) design has scrutinized the network security with several security matrices. Three various situations like health care, smart home, and environment monitoring were selected. A security decision maker decides the major susceptible part of the network. Depending upon the decision taken, the defense operation in opposition to the cyber-attack was confirmed.

In 2017, Fernando A. Teixeira et al. [34] focused the distributed body as a single body and accordingly modeled a widespread program for exchanging information. The program protects the system from attacks caused due to buffer overflow by verifying the information from several nodes. An algorithm was outlined to efficiently discover inter-program associations. They confirmed that the scheme introduced was secured, and the system was designed accurately with appropriate termination. The solution involves minimized runtime of 6% with an average overhead size of the program. The outcome had increased precision, and Tainted Flow Analysis (TFA) selects the memory, which has to be chosen to regulate the overflow.

2.2. Review

Numerous research contributions have been done in the literature to address the issue of security protocols in IoT as shown by Table 1. In [16], integrated privacy preservation scheme has been developed for cybersecurity and privacy threats. Nonetheless, it affects form compatibility and complexity. Also, the ADLS [30] provide better computational cost, and it is efficient for locating the user with less probability. However, it does not reduce the bandwidth usage. Conversely, the semantic model [24] has also been reported in the literature. It has some benefits such as scalability, interoperability, auto arresting and eliminates human errors. Nevertheless, it suffers from some drawbacks such as limited ontologies, and it limits the reusability options. The Game Theory [20] eliminates the loss of objectivity and also it eliminates the security vulnerabilities but to overcome energy efficiency problems is difficult. In [33] Multi-layer cloud architectural model has been adopted for eliminates the heterogeneity issues yet, it lacks from handling global standards. On the other hand, the IoT and cloud computing technology, FIDO protocol [6] is known for its secured transaction. Nevertheless, it does not support if there is a loss in the network packet. In [8], the HARM-based technique has been developed to provide a suitable defense mechanism. Nonetheless, it cannot compute all the potential attacks. Also, [34] proposes the distributed system analysis it protect against buffer overflow attack. However, the main issue of this technique is program insecurity when integer overflow occurs. These drawbacks have extremely motivated for handling privacy preservation mechanism in IoT.

Table 1
Review on security protocols in IoT

Author [citation] Adopted methodology Features Challenges

Jayaraman et al. [16] Integrated privacy preservation scheme ∙ Cyber security
∙ Privacy threats ∙ Compatibility
∙ Complexity

Sun et al. [30] ADLS algorithm ∙ Better computational cost ∙ The efficiency with less probability of locating the user ∙ It does not reduce the bandwidth usage

Mohsin et al. [24] Semantic model ∙ Scalability
∙ Interoperability, auto arresting
∙ Eliminates human errors ∙ Limited ontologies
∙ It limits the reusability options

Seokcheol Lee et al. [20] Game Theory ∙ Eliminates the loss of objectivity
∙ Cost efficiency
∙ Eliminates security vulnerabilities ∙ To overcome energy efficiency problems is difficult

Ming Tao et al. [33] Multi-layer cloud architectural model ∙ Effective and secured home services
∙ Eliminates heterogeneity issues ∙ Lack of handling global standards

Bogdan-Cosmin Chifor et al. [6] FIDO protocol ∙ Secured transaction ∙ Not support if there is a loss in the network packet

Mengmeng Ge et al. [8] HARM-based technique ∙ It provides a suitable defense mechanism ∙ Does not have the ability to compute all the potential attacks

Fernando A. Teixeira et al. [34] Distributed system analysis ∙ Protect against buffer overflow
attack ∙ Program insecurity when integer overflow occurs

Author [citation]	Adopted methodology	Features	Challenges
Jayaraman et al. [16]	Integrated privacy preservation scheme	∙ Cyber security ∙ Privacy threats	∙ Compatibility ∙ Complexity
Sun et al. [30]	ADLS algorithm	∙ Better computational cost ∙ The efficiency with less probability of locating the user	∙ It does not reduce the bandwidth usage
Mohsin et al. [24]	Semantic model	∙ Scalability ∙ Interoperability, auto arresting ∙ Eliminates human errors	∙ Limited ontologies ∙ It limits the reusability options
Seokcheol Lee et al. [20]	Game Theory	∙ Eliminates the loss of objectivity ∙ Cost efficiency ∙ Eliminates security vulnerabilities	∙ To overcome energy efficiency problems is difficult
Ming Tao et al. [33]	Multi-layer cloud architectural model	∙ Effective and secured home services ∙ Eliminates heterogeneity issues	∙ Lack of handling global standards
Bogdan-Cosmin Chifor et al. [6]	FIDO protocol	∙ Secured transaction	∙ Not support if there is a loss in the network packet
Mengmeng Ge et al. [8]	HARM-based technique	∙ It provides a suitable defense mechanism	∙ Does not have the ability to compute all the potential attacks
Fernando A. Teixeira et al. [34]	Distributed system analysis	∙ Protect against buffer overflow attack	∙ Program insecurity when integer overflow occurs

3. Proposed data privacy preservation model

3.1. Adopted architecture

The sensitive data which has to be uploaded to IoT involves two major processes namely, data hiding and data restoration. Initially, for hiding the preserved data, a key should be generated, by which the sensitive data could be encrypted in a secure way. Moreover, the generated key has to be converted into its binary value, which should be similar to the length of the data. Moreover, the main contribution of this paper is to tune or optimize the best key that is most suited for the data sanitization and restoration. The optimization concept with the proposed IDA is used for optimizing the keys which are given randomly. The binary converted optimized key is utilized by the sanitization process, and the corresponding inverse key is utilized for the restoration process. The sanitized (encrypted) sensitive data is subsequently conveyed over the transmission line in a secured manner. The authorized person at the receiver could then obtain the sanitized data using the inverse key. The generated key should be designed optimally so that the correlation between the restored and original data seems too high. The overall architecture of the suggested data preservation model is given in Fig. 1.

Fig. 1.

The overall framework of the proposed data privacy preservation in IoT network.

3.2. Data hiding and restoration

Data hiding: The sensitive data transmitted over the IoT network has to be hidden using key generation before passing it over the transmission line. Prior to the hiding process, the key should be converted to the binary form. The formation of binary data is a specific process that could be carried out as follows: Let the data size be $D_{1} \times D_{2}$ $(200 \times 4)$ , where $D_{1}$ be the records and $D_{2}$ be the fields. The optimal size key could be considered as $20 \times 1$ that is to be multiplied with original sensitive data to produce the sanitized (hidden) data. The key to be processed should be converted into its binary form initially. For that, the elements in $(20 \times 1)$ are partitioned into five subsets, where all the subsets contain 4 elements and every element in the key is transformed into 40 binary bits, such that each subset achieves $(40 \times 4)$ data. Thus the partitioned set contains five $(40 \times 4)$ data, which is said to be the generated binary data. The entire five $(40 \times 4)$ data is concatenated to get the cumulative binary data that would be in size $(200 \times 4)$ . After the production of binary data, it is multiplied by original sensitive data to attain the sanitized data. Figure 2 demonstrates the overall architecture of the data hiding process.

Fig. 2.

The process involved in data hiding scheme.

Data restoration: The generated key comprises of two major components namely, index and sensitive data. In the data restoration process, a vector of the sanitized data of equivalent length is produced for sensitive data that is subsequently multiplied with the optimal key index. Consequently, the multiplied data is added with sanitized data to obtain original or restored data. If the optimized key produced by IDA is accurate, then original data is recovered efficiently. In a further case, if the key produced is not optimal, then the restoration of original data could not be made proficiently. As a result, in this paper, the correlation coefficient of both recovered, and original data is determined to reveal the efficiency of the implemented IDA scheme. The process involved in the data restoration process is given in Fig. 3.

Fig. 3.

Process involved in data restoration scheme.

3.3. Exemplary form

Let us consider that the data could be 5, 3, 3 and 2, where ‘3’ is considered as the sensitive data. The key generated would be 1, 0, 0 and 1. On multiplying the sensitive data and the generated key (after converting into its binary form), the sanitized data could be attained as 5, 0, 0 and 2. Further, by deploying the inverse key, the original data can be accessed as 5, 3, 3, 2. The overall illustration of data hiding and data restoration process is given in Fig. 4.

Fig. 4.

An illustration of data hiding and restoration process.

3.4. Objective model

The objective of the proposed privacy preservation model in IoT network is to generate an optimal key for obtaining the accurate restored data. The objective model L can be formulated as given as in Eq. (1), where S is the sanitized data, O is the original data and R is the data which has to be preserved and N is the number of data and the value of H is given by Eq. (2). $\begin{array}{l} (1) & L = Min (H), \\ (2) & H = Min (\frac{\sum_{i = 1}^{N} S}{\sum_{i = 1}^{N} O} - [\frac{\sum_{i = 1}^{N} O - \sum_{i = 1}^{N} R}{\sum_{i = 1}^{N} O}]) . \end{array}$

4. Optimal key generation based on the DA algorithm

4.1. Solution encoding

The keys to be sanitized are given to the DA algorithm for solution encoding. The number of keys ranging from key 1 to $M_{k}$ is optimized using the DA algorithm, and the better key is identified. The solution encoding process is illustrated in Fig. 5. Here, key (chromosome) is given as solution and the length of the chromosome is given by the Eq. (3). It means the random keys are considered as a solution to the IDA method. $\begin{matrix} (3) & Length of chromosome = \frac{D_{1}}{40} \times D_{2} . \end{matrix}$

Moreover, the boundary limit of chromosome length lies between 1 to $2^{n} - 1$ , where 1 is the minimum boundary limit, and $2^{n} - 1$ is the maximum boundary limit.

Fig. 5.

Keys for solution encoding.

4.2. Improved Dragonfly Algorithm

In nature, dragonflies are measured as tiny predators that hunt all the other minute insects. The major inspiration of the DA technique [15] instigates from static and dynamic swarming behaviors. These two swarming behaviors are found to be incredibly related to the two major phases of optimization using meta-heuristics: exploitation and exploration. Dragonflies generate sub-swarms and fly around several areas in a static swarm that is considered as the major intention of the exploration phase. Accordingly, in the static swarm, nevertheless, dragonflies fly in larger swarms and in a single direction that is constructive in the exploitation phase. All such behaviors are numerically designed as follows.

The separation formulation is measured as in Eq. (4), where $X_{j}$ reveals the jth position of the neighbouring individual, X is the position of the current individual and N is the number of neighbouring individuals. $\begin{matrix} (4) & P_{i} = - \sum_{j = 1}^{N} X - X_{j} . \end{matrix}$

Alignment is evaluated as given by Eq. (5), where $X_{j}$ indicates the velocity of jth neighbouring individual. Moreover, the cohesion formulation is evaluated by Eq. (6), in which $X_{j}$ reveals the position jth neighbouring individual, N represents the number of neighborhoods and X denotes the position of the current individual. $\begin{array}{l} (5) & B_{i} = \frac{\sum_{j = 1}^{N} Q_{j}}{N}, \\ (6) & A_{i} = \frac{\sum_{j = 1}^{N} X_{j}}{N} - X . \end{array}$

Attraction to a food resource is measured by Eq. (7), in which $X^{+}$ exposes the position of the food source and X indicates the position of the current individual. $\begin{matrix} (7) & E_{i} = X^{+} - X . \end{matrix}$

Distraction outwards an enemy is given by Eq. (8), in which $X^{-}$ exposes the position of the enemy and X is the position of the current individual. $\begin{matrix} (8) & F_{i} = X^{-} + X . \end{matrix}$

For updating the artificial dragonflies position in an exploration space and execute their movements, two vectors were taken into account: position (X) and step ( $Δ X$ ).

The step vector reveals the direction of the dragonfly movement as provided in Eq. (9), in which $P_{i}$ denotes the separation of ith individual, p reveals the separation weight, a is the alignment weight and A represents the ith individual cohesion, c signifies the cohesion weight, B exposes the alignment of i individual, $E_{i}$ signifies the food resource of the ith individual, f denotes the food factor, e indicates the enemy factor, w provides the inertia weight and $F_{i}$ symbolizes the position of enemy of ith individual t signifies the iteration counter. $\begin{matrix} (9) & Δ X_{t + 1} = (p P_{i} + a B_{i} + c A_{i} + f E_{i} + e F_{i}) + w Δ X_{t} . \end{matrix}$

After the evaluation of step vector, the position vectors are formulated as in Eq. (10), in which t denotes the current iteration. $\begin{matrix} (10) & X_{t + 1} = X_{t} + Δ X_{t + 1} . \end{matrix}$

For enhancing the stochastic behavior, exploration, and randomness of the artificial dragonflies, they are necessitated to fly around the exploration space using an arbitrary walk, while there is an absence of neighboring solutions. Under such situations, the dragonfly position is updated by Eq. (11), in which z indicates the dimension of the position vectors and t denotes the current iteration. $\begin{matrix} (11) & X_{t + 1} = X_{t} + Levy (z) \times X_{t} . \end{matrix}$

The Levy flight is measured as in Eq. (12), in which, β is a constant, $r_{1}$ and $r_{2}$ are two arbitrary numbers that are lying between $[0, 1]$ , and δ can be evaluated as in Eq. (13), in which $Γ (x) = (x - 1)$ . $\begin{array}{l} (12) & Levy (x) = 0.01 \times \frac{r_{1} \times δ}{| r_{2} |^{\frac{1}{β}}}, \\ (13) & δ = {(\frac{Γ (1 + β) \times sin (\frac{π β}{2})}{Γ \frac{(1 + β)}{2} \times β \times 2^{(\frac{β - 1}{2})}})}^{\frac{1}{β}} . \end{array}$

The pseudo code of the conventional DA scheme is given by Algorithm 1.

Algorithm 1
Conventional Dragonfly optimization for optimal key generation

Initialize the population of dragonfly $X_{i}$ ( $i = 1, 2, \dots, n$ )

Initialize step vectors $Δ X_{i}$ ( $i = 1, 2, \dots, n$ )

while the end condition is not fulfilled

Compute the objective value of entire fireflies

Update enemy and food source

Update w, s, a, c, f and e

Compute P, B, A, F and E using Eq. (4)–(8)

Neighboring radius is updated

if a dragonfly involves one neighbor dragonfly,

Update velocity vector using Eq. (9)

Update position vector using Eq. (10)

else

Update position vector using Eq. (11)

end if

Verify and approve the novel positions depending on variable boundaries

end while

4.3. Proposed Dragonfly Algorithm

To obtain an optimal key for secure sanitization and restoration of sensitive data, improved DA is adopted. The conventional DA algorithm could be able to solve only continuous and single objective optimization problems. Hence, this paper adopts a self-improved DA to overcome the challenges prevailing in it. The best dragonfly is denoted by $X_{B}$ , and the worst dragonfly is denoted by $X_{W}$ . According to the improved DA, after updating w, s, a, c, f and e, the values of P, B and A using Eq. (4)–(6) are only evaluated.

In improved DA algorithm, not only the food position is defined, but also the fitness position is also described and compared with the threshold fitness, as given by Eq. (14). $\begin{matrix} (14) & Fitness threshold = food fitness + (food fitness \times 0.2) . \end{matrix}$

Therefore, the attraction of $X_{B}$ towards the food will be obtained as a minimum the distraction of $X_{B}$ from enemy would be maximum. Subsequently, the neighboring radius r is updated. Under a condition that the dragonfly includes a neighbor, velocity is updated using Eq. (6) and Eq. (7). Finally, update the positions of the dragonflies depending on the optimal position and distance. The pseudo code of the proposed IDA approach is given by Algorithm 2.

Algorithm 2
Proposed dragonfly optimization optimal key generation

Initialize the population of dragonfly $X_{i}$ ( $i = 1, 2, \dots, n$ )

Initialize step vectors $Δ X_{i}$ ( $i = 1, 2, \dots, n$ )

while the end condition is not fulfilled

Compute the objective value of entire fireflies

Update enemy and food source

Update w, s, a, c, f and e

Compute P, B and A using Eq. (4)–(6)

If Distance from $X_{B}$ th position to food position < r and
Distance from $X_{B}$ fitness to food fitness ⩾ threshold

Compute E using Eq. (7)

Else

$E = zeros (d, 1)$

If Distance from $X_{B}$ th position to enemy position < r and
Distance from $X_{B}$ th fitness to enemy fitness ⩾ threshold

Compute F using Eq. (8)

Else

$F = zeros (d, 1)$

Neighboring radius is updated

if a dragonfly involves one neighbor dragonfly,

Update velocity vector using Eq. (9)

Update position vector using Eq. (10)

else

Update position vector using Eq. (11)

end if

Verify and approve the novel positions depending on variable boundaries

end while

5. Results and discussions

5.1. Simulation procedure

The proposed data sanitization and restoration model in IoT network were simulated in MATLAB 2015a, and the achieved results were observed. The experimentation was conducted using physical activity data monitoring, http://archive.ics.uci.edu/ml/datasets/pamap2+physical+activity+monitoring. It contains 18 different physical activities (such as walking, cycling, playing soccer, etc.), performed by 9 subjects wearing 3 inertial measurement units and a heart rate monitor. Each data is of $[600 \times 4]$ , i.e., 600 records and 4 fields. The synthetic data was produced from the original data. The corresponding data was varied in 10%, 20%, and 30% respectively and generated three test cases: Test case 1, Test case 2 and Test case 3. For every change, arbitrary data were produced, i.e., for 10% variation random data were generated in the range of (−10 to +10), for 20% variation, data were created in the range of (−20 to +20), for 30% variation, data were created in the range of (−30 to +30) either by adding or subtracting the value. Therefore, for every test case, there will be ten synthetic data. Next, to the implementation, the proposed method was compared with conventional schemes like GA [36], PSO [18], ABC [41], FF [38] and DA [15] and the results were obtained. In the analytical part, the analysis corresponding to convergence, sanitization potency, restoration potency, key sensitivity, and different attacks has been focussed. Here, the performance of sanitization and restoration is better if the cost function is minimum, and restoration efficiency is maximum.

5.2. Convergence analysis

The convergence analysis for the proposed model for obtaining accurate data sanitization and data restoration process over IoT network concerning iteration is given in Fig. 5. From Fig. 6(a), the convergence analysis for Test case 1 for the suggested model is 86.6% better than GA, 78.9% better than ABC, 80% better than PSO, 86.2% better than FF and 50% better than DA algorithms. Similarly, the convergence analysis for Test case 2 can be achieved from Fig. 6(b), where the suggested model is 50% superior to GA, 73.6% superior to ABC, 68.75% superior to PSO, 81.4% superior to FF and 20% superior to DA algorithms. Also, from Fig. 6(c), the implemented scheme for Test case 3 is 89.2% better than GA, 81.25% better than ABC, 81.25% better than PSO, 80% better than FF and 50% better than DA algorithms. Thus the capability of the suggested algorithm in revealing better convergence analysis has been proved.

Fig. 6.

Convergence analysis for the proposed and conventional algorithms: (a) Test case 1, (b) Test case 2, (c) Test case 3.

5.3. Sanitization potency

The sanitization analysis for the suggested model is given in Fig. 7 for ten experiments. From Fig. 7(a), the sanitization analysis for Experiment I for Test case 1 can be obtained, where the suggested model is 40% superior to GA, 20% superior to ABC, 20% superior to PSO, 78.5% superior to FF and 26.6% superior to DA algorithm. Moreover, the proposed sanitization scrutiny for Experiment II for Test cases 1 from Fig. 7(b) is 60% better than GA, 30% better than ABC, 40% better than PSO, 85.5% better than FF and 75% better than DA techniques. Also, from Fig. 7(c), the sanitization study for Experiment III for Test case 1 is noted, where the proposed model is 75% superior to GA, 71.4% superior to ABC, 66.6% superior to PSO, 84% superior to FF and 20% superior to DA schemes. Further, from Fig. 7(d), the sanitization analysis for Experiment IV for Test case 3 can be achieved, where the presented model is 86.5% superior to GA, 73.3% superior to ABC, 50% superior to PSO, 75% superior to FF and 20% superior to DA methods. Similarly, from Fig. 7(e), the sanitization study for Experiment V for Test case 2 is observed, in which the proposed model is 13% better than GA, 70% better than ABC, 60% better than PSO, 92.3% better than FF and 10% better than DA schemes. From Fig. 7(f), the sanitization analysis for Experiment VI for Test case 1 is attained, where the introduced algorithm is 84.16% superior to GA, 72.8% superior to ABC, 68.3% superior to PSO, 81.9% superior to FF and 5.26% superior to DA methods. Moreover, the presented model for Experiment VII for Test case 3 can be achieved from Fig. 7(g), which is 77.7% better than GA, 20% better than ABC, 20% better than PSO, 76.9% better than FF and 20% better than DA schemes. From Fig. 7(h), the sanitization study for Experiment VIII for Test case 2 is represented, in which the suggested model is 75% superior to GA, 30% superior to ABC, 30% superior to PSO, 84% superior to FF and 20% superior to DA schemes. Also, from Fig. 7(i), the sanitization study for Experiment IX for Test case 3 is measured, in which the implemented method is 83.2% better than GA, 70.58% better than ABC, 72.2% better than PSO, 81.48% better than FF and 3.8% better than DA algorithms. Finally, the sanitization study for Experiment X for Test case 1 From Fig. 7(j) is pointed out, in which the suggested method is 78.5% superior to GA, 20% superior to ABC, 70% superior to PSO, 78.5% superior to FF and 26.6% superior to DA techniques. Thus the enhancement of the proposed algorithm regarding sanitization analysis has been confirmed.

Fig. 7.

Sanitization analysis for the proposed and conventional algorithm for Test case 1, Test case 2, Test case 3: (a) Experiment I, (b) Experiment II, (c) Experiment III, (d) Experiment IV, (e) Experiment V, (f) Experiment VI, (g) Experiment VII, (h) Experiment VIII, (i) Experiment IX, (j) Experiment X.

5.4. Restoration potency

The restoration analysis for the proposed model is specified in Fig. 8 for three cases for ten experiments. From Fig. 8(a), the restoration analysis for an Experiment I for Test case 1 is noted, where the proposed model is 9% superior to ABC, 10.1% superior to GA, 4% superior to DA, 10.1% superior to FF and 9.5% superior to PSO algorithm. Also, from Fig. 8(b), the suggested restoration analysis for Experiment II for Test case 1 is 15.5% better than FF, 14.14% better than GA, 9% better than PSO, 10.1% better than ABC and 1% better than DA techniques. As well, from Fig. 8(c), the restoration study for Experiment III for Test case 3 can be obtained, where the proposed model is 17.17% superior to GA, 16.16% superior to PSO, 17.17% superior to FF, 15.15% superior to ABC and 4% superior to DA schemes. Similarly, from Fig. 8(d), the restoration analysis for Experiment IV for Test case 2 can be predicted, where the presented model is 14.14% superior to ABC, 16.16% superior to GA,14.14% superior to PSO, 1% superior to DA and 16.16% superior to FF methods. Similarly, from Fig. 8(e), the restoration study for Experiment V for Test case 1 is noticed, in which the implemented model is 6.25% better than GA, 8.33% better than FF, 3.12% better than PSO, 4.16% better than ABC and 1.04% better than DA schemes. From Fig. 8(f), the restoration analysis for Experiment VI for Test case 3 is obtained, in which the suggested algorithm is 17.17% superior to GA, 16.16% superior to PSO, 16.16% superior to ABC, 17.17% superior to FF and 1% superior to DA methods. Furthermore, the presented algorithm for Experiment VII for Test case 1 can be achieved from Fig. 8(g), which is 12.12% better than GA, 8% better than ABC, 8% better than PSO, 12.12% better than FF, and 1% better than DA schemes. From Fig. 8(h), the restoration study for Experiment VIII for Test case 1 is represented, in which the suggested model is 11.11% superior to GA, 9% superior to ABC, 10.1% superior to PSO, 12.12% superior to FF and 4% superior to DA schemes. Also, from Fig. 8(i), the restoration study for Experiment IX for Test case 3 is pointed out, in which the implemented method is 16.16% better than ABC, 19.19% better than GA, 15.15% better than PSO and 19.19% better than FF algorithms. At last, the restoration study for Experiment X for Test case 1 is measured from Fig. 8(j), in which the suggested method is 9% superior to ABC, 12.12% superior to PSO 10.1% superior to GA, and 10.1% superior to FF techniques. Thus, the development of the implemented privacy preservation algorithm concerning restoration analysis has been established.

Fig. 8.

Convergence analysis for the proposed and conventional algorithms for Test case 1, Test case 2, Test case 3: (a) Experiment I, (b) Experiment II, (c) Experiment III, (d) Experiment IV, (e) Experiment V, (f) Experiment VI, (g) Experiment VII, (h) Experiment VIII, (i) Experiment IX, (j) Experiment X.

5.5. Key sensitivity analysis

The key sensitivity analysis for the proposed data sanitization and data restoration model for 10%, 30%, 40%, and 70% is specified by Table 2, Table 3 and Table 4 for three cases for ten experiments. From the Table 2, for Test case 1, it can be noted that, for 10%, the proposed method is 5.35% better than GA, 1.76% better than ABC, 1.71% better than PSO, 3.35% better than FF and 2.36% better than DA schemes. Similarly, for 30%, the suggested model is 1.2% superior to GA, 5.27% superior to ABC, 2.83% superior to PSO, 5.41% superior to FF and 3.91% superior to DA methods. Similarly, from Table 3, for Test case 2, the proposed key sensitivity for 30% is 6.85% better than GA, 1.14% better than ABC, 5.38% better than PSO, 2.88% better than FF and 8.02% better than DA schemes. Similarly, for 70%, the presented model is 3.28% superior to GA, 6.28% superior to ABC, 8.13% superior to PSO, 0.9% superior to FF and 5.99% superior to DA methods. Also, from Table 4, for Test case 3, the implemented model for 30% is 5.17% better than GA, 6.77% better than ABC, 4.63% better than PSO, 4.37% better than FF and 7.44% better than DA schemes. Likewise, for 40%, the implemented scheme is 4.33% superior to GA, 0.15% superior to ABC, 1.23% superior to PSO, 3.52% superior to FF and 2.98% superior to DA schemes. Thus the enhanced key sensitivity with better restoration capability was found to be exhibited using the proposed approach.

5.6. Statistical analysis

As the statistical analysis is met heuristic in nature, the proposed scheme was iterated several times, and the better result was taken. The statistical analysis for the proposed model is specified by Table 5, Table 6, and Table 7 for three cases. For Test case 1, the best performance of the proposed method is 73.9% superior to GA, 73.24% superior to ABC, 73.63% superior to PSO, and 70.23% superior to FF and 32.8% superior to DA techniques. Further, the worst performance of the suggested scheme is 71.5% better than GA, 71% better than ABC, 71.8% better than PSO, 70.22% better than FF and 42.04% better than DA schemes. Similarly, for Test case 2, the presented method for mean estimation is 81.57% superior to GA, 80.92% superior to ABC, 81.25% superior to PSO, and 80.33% superior to FF and 35.15% superior to DA techniques. Also, the median performance of the suggested model is 83.1% better than GA, 82.86% better than ABC, 82.23% better than PSO, 82.3% better than FF and 44.1% better than DA methods. Similarly, for Test case 3, the implemented model regarding the best performance is 73.9% superior to GA, 73.24% superior to ABC, 73.63% superior to PSO, and 70.23% superior to FF and 32.8% superior to DA techniques. Moreover, for Test case 3, the proposed method regarding best performance is 92.5% superior to GA, 92.6% superior to ABC, 92.7% superior to PSO, and 92.2% superior to FF and 50.4% superior to DA techniques. Also, the standard deviation for the proposed model is 51.4% superior to GA, 56.9% superior to ABC, 34.7% superior to PSO, and 35.9% superior to FF and 59.3% superior to DA techniques. Thus the enhancement of the implemented model regarding statistical analysis has been confirmed.

Table 2
Key sensitivity analysis for proposed and conventional methods for Test case 1

Methods 10% 30% 40% 70%

GA [36] 0.793428 0.74664 0.811702 0.643115

ABC [18] 0.832226 0.776299 0.808966 0.652855

PSO [41] 0.84975 0.758177 0.797185 0.628397

FF [38] 0.835347 0.777358 0.773513 0.622441

DA [15] 0.815181 0.766287 0.807078 0.603754

IDA 0.834724 0.737115 0.755306 0.542988

Methods	10%	30%	40%	70%
GA [36]	0.793428	0.74664	0.811702	0.643115
ABC [18]	0.832226	0.776299	0.808966	0.652855
PSO [41]	0.84975	0.758177	0.797185	0.628397
FF [38]	0.835347	0.777358	0.773513	0.622441
DA [15]	0.815181	0.766287	0.807078	0.603754
IDA	0.834724	0.737115	0.755306	0.542988

Table 3

Key sensitivity analysis for proposed and conventional methods for Test case 2

Methods	10%	30%	40%	70%
GA [36]	0.919385	0.867836	0.729531	0.724231
ABC [18]	0.914688	0.84596	0.693295	0.745334
PSO [41]	0.930724	0.879129	0.719481	0.758585
FF [38]	0.911527	0.840301	0.701927	0.694579
DA [15]	0.928091	0.869447	0.737909	0.743465
IDA	0.906184	0.818554	0.682226	0.700958

Table 4

Key sensitivity analysis for proposed and conventional methods for Test case 3

Methods	10%	30%	40%	70%
GA [36]	0.951551	0.787232	0.772462	0.813496
ABC [18]	0.942405	0.799966	0.741685	0.796677
PSO [41]	0.941663	0.783371	0.749087	0.815169
FF [38]	0.94828	0.781319	0.766292	0.802601
DA [15]	0.94916	0.804601	0.762308	0.817447
IDA	0.941979	0.748288	0.739892	0.788792

5.7. Comparison based on attacks

The attacks like Known Plain Text Attacks (KPA) and Cipher Plain Text Attacks (CPA) are defined in this section. KPA is analyzed by correlating one original data with all original data and one sanitized data with all sanitized data. Likewise, the CPA analysis is defined by correlating each sanitized data with its corresponding restored data. The proposed model regarding CPA attack is given by Table 8 for three test cases. From Table 8, the suggested model for Test case 1 is 1.64% superior to GA, 2.95% superior to ABC, 1.14% superior to PSO, and 5.76% superior to FF and 2.24% superior to DA techniques. Similarly, for Test case 2, the proposed scheme is 5% better than GA, 5.93% better than ABC, 5.72% better than PSO, 2.96% better than FF and 0.2% better than DA methods. On considering KPA attack from Table 9, the implemented model for Test case 1 is 0.8% superior to GA, 0.3% superior to ABC, 0.1% superior to PSO, 1.7% superior to FF and 0.5% superior to DA schemes. Also, for Test case 3, the presented method is 5.6% better than GA, 3.38% better than ABC, 1.57% better than PSO, 1.7% better than FF and 5.19% better than DA methods. Thus the enhanced computation of the proposed method has been validated by the simulation results.

Table 5
Statistical analysis of proposed and conventional methods for Test case 1

Methods GA [36] ABC [18] PSO [41] FF [38] DA [15] IDA

Best 1.190964 1.176592 1.202085 1.099208 0.231511 0.151396

Worst 1.45757 1.434373 1.471775 1.393836 0.716034 0.415631

Mean 1.278184 1.240223 1.270788 1.188576 0.390132 0.245661

Median 1.247977 1.201845 1.22253 1.150868 0.323809 0.209263

Standard deviation 0.103602 0.109218 0.113444 0.116914 0.195994 0.100742

Methods	GA [36]	ABC [18]	PSO [41]	FF [38]	DA [15]	IDA
Best	1.190964	1.176592	1.202085	1.099208	0.231511	0.151396
Worst	1.45757	1.434373	1.471775	1.393836	0.716034	0.415631
Mean	1.278184	1.240223	1.270788	1.188576	0.390132	0.245661
Median	1.247977	1.201845	1.22253	1.150868	0.323809	0.209263
Standard deviation	0.103602	0.109218	0.113444	0.116914	0.195994	0.100742

Table 6

Statistical analysis of proposed and conventional methods for Test case 2

Methods	GA [36]	ABC [18]	PSO [41]	FF [38]	DA [15]	IDA
Best	1.190964	1.176592	1.202085	1.099208	0.231511	0.151396
Worst	1.45757	1.434373	1.471775	1.393836	0.716034	0.415631
Mean	1.278184	1.240223	1.270788	1.188576	0.390132	0.245661
Median	1.247977	1.201845	1.22253	1.150868	0.323809	0.209263
Standard deviation	0.103602	0.109218	0.113444	0.116914	0.195994	0.100742

Table 7

Statistical analysis of proposed and conventional methods for Test case 3

Methods	GA [36]	ABC [18]	PSO [41]	FF [38]	DA [15]	IDA
Best	1.190964	1.176592	1.202085	1.099208	0.231511	0.151396
Worst	1.45757	1.434373	1.471775	1.393836	0.716034	0.415631
Mean	1.278184	1.240223	1.270788	1.188576	0.390132	0.245661
Median	1.247977	1.201845	1.22253	1.150868	0.323809	0.209263
Standard deviation	0.103602	0.109218	0.113444	0.116914	0.195994	0.100742

Table 8

Analysing KPA attack on proposed and conventional models

Methods	Test case 1	Test case 2	Test case 3
GA	0.982931	0.990231	0.984528
ABC	0.967486	0.998346	0.996991
PSO	0.985628	0.996425	0.995911
FF	0.939167	0.974803	0.971352
GSO	0.974243	0.944145	0.978147
IDA	0.996397	0.942089	0.955475

Table 9

Analysing CPA attack on proposed and conventional models

Methods	Test case 1	Test case 2	Test case 3
GA	0.987476	0.988491	0.993106
ABC	0.976729	0.982445	0.972847
PSO	0.980453	0.989616	0.955869
FF	0.996044	0.978997	0.941126
GSO	0.97408	0.964921	0.989258
IDA	0.97896	0.941977	0.940164

6. Conclusion

This paper has presented the improved data sanitization and restoration model over the IoT network. For hiding the preserved data, a key was generated, by which the sensitive data was sanitized in a secure way. Moreover, the generated key was converted into its binary value, which should be similar to the length of the data. Then the sanitized (encrypted) sensitive data was subsequently conveyed over the transmission line in a secured manner. The authorized person at the receiver side could then obtain the sanitized data using the inverse key. For better key generation, recently established optimization technique called Dragonfly Algorithm (DA) was improved and adopted to attain the objective model. Moreover, the suggested method was compared with various conventional algorithms, and the results were obtained. From the analysis of sanitization study for Experiment III for Test case 1, the proposed model is 75% superior to GA, 71.4% superior to ABC, 66.6% superior to PSO, 84% superior to FF and 20% superior to DA schemes. On considering the restoration analysis, the suggested model for Experiment II for Test case 1 is 60% better than GA, 30% better than ABC, 40% better than PSO, 85.5% better than FF and 75% better than DA techniques. Thus the improvement of the proposed algorithm was confirmed by various research analyses in preserving data in IoT.

References

Acemoglu,

Makhdoumi,

Malekian and

Ozdaglar, Privacy-constrained network formation, Games and Economic Behavior 105 (2017), 255–275. doi:10.1016/j.geb.2017.08.001.

Asplund and

Nadjm-Tehrani, Attitudes and perceptions of IoT security in critical societal services, IEEE Access 4 (2016), 2130–2138. doi:10.1109/ACCESS.2016.2560919.

Catuogno and

Turchi, The dark side of the interconnection: Security and privacy in the web of things, in: 9th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, 2015. doi:10.1109/imis.2015.86.

Chen,

C.E.

Beaudoin and

Hong, Securing online privacy: An empirical test on Internet scam victimization, online privacy concerns, and privacy protection behaviors, Computers in Human Behavior 70 (2017), 291–302. doi:10.1016/j.chb.2017.01.003.

Chen

et al., Robustness, security and privacy in location-based services for future IoT: A survey, IEEE Access 5 (2017), 8956–8977. doi:10.1109/ACCESS.2017.2695525.

B.-C.

Chifor,

Bica,

V.-V.

Patriciu and

Pop, A security authorization scheme for smart home Internet of Things devices, Future Generation Computer Systems 86 (2018), 740–749. doi:10.1016/j.future.2017.05.048.

Fiorentini ,

Rossi ,

Bernardeschi ,

Cantore and

Guadagni , Is there a new drug beyond floxuridine for intra-arterial hepatic chemotherapy in liver metastases from colorectal cancer?, Journal of Clinical Oncology 23(9) (2015), 2105. doi:10.1200/JCO.2005.99.297.

Ge,

J.B.

Hong,

Guttmann and

D.S.

Kim , A framework for automating security analysis of the Internet of Things, Journal of Network and Computer Applications 83 (2017), 12–27. doi:10.1016/j.jnca.2017.01.033.

Giuliano,

Mazzenga,

Neri and

A.M.

Vegni, Security access protocols in IoT capillary networks, IEEE Internet of Things Journal 4(3) (2017), 645–657. doi:10.1109/JIOT.2016.2624824.

10.

Guadagni,

Fiorentini,

Clementi,

Palumbo,

Masedu,

Deraco,

De Manzoni,

Chiominto,

Valenti and

Pellegrini, MGMT methylation correlates with melphalan pelvic perfusion survival in stage III melanoma patients. A pilot study, Melanoma Research 27 (2017), 439–447. doi:10.1097/CMR.0000000000000367.

11.

Guadagni,

Russo,

Abate,

Pozone,

Capannolo,

Marsili,

D’Alessandro,

Amicucci and

K.R.

Aigner, Stop-flow in mediastinum and thorax for resistant lymphoma, Hepato-gastroenterology 47 (2000), 378–382.

12.

Henze,

Hermerschmidt,

Kerpen,

Haußling,

Rumpe and

Wehrle , A comprehensive approach to privacy in the cloud-based, Internet of Things, Future Generation Computer Systems, 56 (2016), 701–718. doi:10.1016/j.future.2015.09.016.

13.

M.S.

Hossain,

Muhammad,

S.M.M.

Rahman,

Abdul,

Alelaiwi and

Alamri, Toward end-to-end biometrics-based security for IoT infrastructure, IEEE Wireless Communications 23(5) (2016), 44–51. doi:10.1109/MWC.2016.7721741.

14.

Jacobsson,

Boldt and

Carlsson, A risk analysis of a smart home automation system, Future Generation Computer Systems 56 (2016), 719–733. doi:10.1016/j.future.2015.09.003.

15.

Jafari,

Hossein and

Chaleshtari, Using dragonfly algorithm for optimization of orthotropic infinite plates with a quasi-triangular cut-out, European Journal of Mechanics – A/Solids 66 (2017), 1–14. doi:10.1016/j.euromechsol.2017.06.003.

16.

P.P.

Jayaraman ,

Yang,

Yavari,

Georgakopoulos and

Yi, Privacy preserving Internet of Things: From privacy techniques to a blueprint architecture and efficient implementation, Future Generation Computer Systems 76 (2017), 540–549. doi:10.1016/j.future.2017.03.001.

17.

Kang,

Kim and

Choo, Internet of everything: A large-scale autonomic IoT gateway, IEEE Transactions on Multi-Scale Computing Systems 3(3) (2017), 206–214. doi:10.1109/TMSCS.2017.2705683.

18.

Koçer, Bollinger bands approach on boosting ABC algorithm and its variants, Applied Soft Computing 49 (2016), 292–312. doi:10.1016/j.asoc.2016.08.023.

19.

Kuusijarvi,

Savola,

Savolainen and

Evesti, Mitigating IoT security threats with a trusted network element, in: 11th International Conference for Internet Technology and Secured Transactions (ICITST), Barcelona, 2016, pp. 260–265.

20.

Lee,

Kim,

Choi and

Shon , Game theory-based security vulnerability quantification for social Internet of Things, Future Generation Computer Systems 82 (2018), 752–760. doi:10.1016/j.future.2017.09.032.

21.

Lopez,

Rios,

Bao and

Wang, Evolving privacy: From sensors to the Internet of Things, Future Generation Computer Systems 75 (2017), 46–57. doi:10.1016/j.future.2017.04.045.

22.

Luo,

Zhang,

L.T.

Yang,

Liu,

Chang and

Ning, A kernel machine-based secure data sensing and fusion scheme in wireless sensor networks for the cyber-physical systems, Future Generation Computer Systems 61 (2016), 85–96. doi:10.1016/j.future.2015.10.022.

23.

Mohsin,

Anwar,

Husari,

Al-Shaer and

M.A.

Rahman, IoTSAT: A formal framework for security analysis of the Internet of Things (IoT), in: IEEE Conference on Communications and Network Security (CNS), Philadelphia, PA, 2016, pp. 180–188.

24.

Mohsin ,

Anwar,

Zaman and

Al-Shaer, IoTChecker: A data-driven framework for security analytics of Internet of Things configurations, Computers & Security 70 (2017), 199–223. doi:10.1016/j.cose.2017.05.012.

25.

R.V.

Ravi and

Subramaniam, Optimized wavelet filters and modified huffman encoding-based compression and chaotic encryption for image data, International Journal of Applied Engineering Research 12(13) (2017), 3961–3977.

26.

Said,

Albagory,

Nofal and

Al Raddady, IoT-RTP and IoT-RTCP: Adaptive protocols for multimedia transmission over Internet of Things environments, IEEE Access 5 (2017), 16757–16773. doi:10.1109/ACCESS.2017.2726902.

27.

Sajid,

Abbas and

Saleem, Cloud-assisted IoT-based SCADA systems security: A review of the state of the art and future challenges, IEEE Access, 4 (2016), 1375–1384.

28.

Sherifi and

Senja , Internet usage on mobile devices and their impact on evolution of informative websites in Albania, European Journal of Business, Economics and Accountancy 3 (2018), 37–43.

29.

Shirvanimoghaddam,

Condoluci,

Dohler and

S.J.

Johnson, On the fundamental limits of random non-orthogonal multiple access in cellular massive IoT, IEEE Journal on Selected Areas in Communications 35(10) (2017), 2238–2252. doi:10.1109/JSAC.2017.2724442.

30.

Sun,

Chang,

Ramachandran,

Sun,

Li,

Yu and

Liao, Efficient location privacy algorithm for Internet of Things (IoT) services and applications, Journal of Network and Computer Applications 89 (2017), 3–13. doi:10.1016/j.jnca.2016.10.011.

31.

Sun,

Chang,

Ramachandran,

Sun and

Liao, Efficient location privacy algorithm for Internet of Things (IoT) services and applications, Journal of Network and Computer Applications 89 (2017), 3–13. doi:10.1016/j.jnca.2016.10.011.

32.

J.C.

Talwana and

H.J.

Hua, Smart world of Internet of Things (IoT) and its security concerns, in: 2016 IEEE International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom) and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData), Chengdu, 2016, pp. 240–245.

33.

Tao ,

Zuo ,

Liu,

Castiglione and

Palmieri, Multi-layer cloud architectural model and ontology-based security service framework for IoT-based smart homes, Future Generation Computer Systems, 78(Part 3) (2018), 1040–1051. doi:10.1016/j.future.2016.11.011.

34.

F.A.

Teixeira,

F.M.Q.

Pereira,

H.-C.

Wong,

J.M.S.

Nogueira and

L.B.

Oliveira, SIoT: Securing Internet of Things through distributed systems analysis, Future Generation Computer Systems, 92 (2019), 1172–1186. doi:10.1016/j.future.2017.08.010.

35.

R.T.

Tiburski,

L.A.

Amaral,

de Matos,

D.F.G.

de Azevedo and

Hessel, The role of lightweight approaches towards the standardization of a security architecture for IoT middleware systems, IEEE Communications Magazine 54(12) (2016), 56–62. doi:10.1109/MCOM.2016.1600462CM.

36.

T.D.

Vrionis,

X.I.

Koutiva and

N.A.

Vovos, A genetic algorithm-based low voltage ride-through control strategy for grid connected doubly fed induction wind generators, IEEE Transactions on Power Systems 29(3) (2014), 1325–1334. doi:10.1109/TPWRS.2013.2290622.

37.

A.M.

Wagh and

S.R.

Todmal, Eyelids, eyelashes detection algorithm and hough transform method for noise removal in iris recognition, International Journal of Computer Applications 112(3) (2015), 28–31. doi:10.5120/19647-1239.

38.

Wang ,

Wang,

Zhou,

Sun and

Cui, Firefly algorithm with neighborhood attraction, Information Sciences 382–383 (2017), 374–387. doi:10.1016/j.ins.2016.12.024.

39.

Xu,

Ren,

Song and

Du, Security enhancement for IoT communications exposed to eavesdroppers with uncertain locations, IEEE Access 4 (2016), 2840–2853. doi:10.1109/ACCESS.2016.2575863.

40.

Yoo and

Shon, Challenges and research directions for heterogeneous cyber physical system based on IEC 61850: Vulnerabilities, security requirements, and security architecture, Future Generation Computer Systems 61 (2016), 128–136. doi:10.1016/j.future.2015.09.026.

41.

Zhang and

Xia, An improved PSO algorithm for parameter identification of nonlinear dynamic hysteretic models, Journal of Sound and Vibration 389 (2017), 153–167. doi:10.1016/j.jsv.2016.11.006.

42.

Zhou,

Cao,

Dong and

A.V.

Vasilakos, Security and privacy for cloud-based IoT: Challenges, IEEE Communications Magazine 55(1) (2017), 26–33. doi:10.1109/MCOM.2017.1600363CM.