IoT device type identification using training deep quantum neural networks optimized with a chimp optimization algorithm for enhancing IoT security

Abstract

IoT networks can be defined as groups of physically connected things and devices that can connect to the Internet and exchange data with one another. Since enabling an increasing number of internets of things devices to connect with their networks, organizations have become more vulnerable to safety issues and attacks. A major drawback of previous research is that it can find out prior seen types only, also any new device types are considered anomalous. In this manuscript, IoT device type detection utilizing Training deep quantum neural networks optimized with a Chimp optimization algorithm for enhancing IOT security (IOT-DTI-TDQNN-COA-ES) is proposed. The proposed method entails three phases namely data collection, feature extraction and detection. For Data collection phase, real network traffic dataset from different IoT device types are collected. For feature mining phase, the internet traffic features are extracted through automated building extraction (ABE) method. IoT device type identification phase, Training deep quantum neural networks (TDQNN) optimized with Chimp optimization algorithm (COA) is utilized to detect the category of IoT devices as known and unknown device. IoT network is implemented in Python. Then the simulation performance of the proposed IOT-DTI-TDQNN-COA-ES method attains higher accuracy as26.82% and 23.48% respectively, when compared with the existing methods.

Keywords

IOT security device type identification training deep quantum neural networks chimp optimization algorithm

1. Introduction

In the twenty-first century a critical need for IOT device safety. IOT connects everyone and everything, but it also opens up a lot of potential for different attack types to succeed [10]. Even though the term “Internet of Things” is brief in its circumstances, it encompasses the whole world with its sophisticated technology and imaginable services. Internet of things is employed to establish a connection amongst human and virtual world under several smart devices and its services via various communication protocols. Through an open network, IoT devices are used for a number of purposes, improving the users’ access to the devices [11]. Although IoT advances technology and makes life easier and more convenient, it also puts consumers’ privacy at danger from various attacks. The safety of IOT devices has become a major concern because anyone can access some IOT devices from anywhere without the user’s permission. To safeguard, numerous security measures have been implemented. But, the physical design limitations their processing, also limits their ability to apply sophisticated security protocols [12]. A threat/attack occurs when an unauthorized user gains access to a system as well as discloses private data without associated user’s consent. Typical computer security system can be separated as 2 types: network and host. Both systems have diverse incorporated security modules, like firewalls, intrusion detection schemes, antivirus software, which observe the system as well as alert while spiteful actions happens [5]. Amongst them, intrusion detection scheme contributes a significant part on information safety technology. This is an essential safety approach to handle network attacks, also detect spiteful actions. Due to the tremendous number of malware produced every day, including worms, viruses, Trojan horses, botnets, ransomware, many attackers still hold a disproportionate amount of power, despite the fact that many companies on the market create embedded safety modules. Additionally, a lot of small- and medium-sized networks are incapable of timely updating virus data bases along patches, or linked devices lack any protection capabilities. These networks have highly susceptible to attackers. Therefore, this is required to design automatic systems to identify with categorize malware. IoT networks are collections of physical items and equipment that may connect to the Internet and exchange data with one another. They are embedding with sensor, actuator, computing, storage, communication mechanisms [18].

The training Deep Quantum Neural Networks to realize the IOT devices linked with network. This is probable to identify distinct configurations that can differentiate device types. The proposed technique can protect the IOT networks’ process effectually against diverse attacks by verifying and examining IOT devices’ process.

The Chimp optimization Algorithm (COA) is employed to enhance the parameter structure of TDQNN and determine the proper hyper parameters to avert the higher labor price of manually tuning the parameters that determine the identification task fit to the network attack category. Chimp Optimization Algorithm optimizes the DQNN’s parameters, making it more robust to variations and noise in the input data. This robustness can help mitigate the challenges posed by the heterogeneity and dynamic nature of IoT environments. The optimized DQNN can better handle variations in device behavior, environmental conditions, and data quality.

The major drawback is that it can only recognize device types that have been seen before and that it needs labeled data for training. Device types that are added later on will be regarded as unknown types. Since there are hundreds of millions of new IoT devices created annually in the real world, a system that relies exclusively on supervised learning would be unworkable.

The main contribution of this manuscript is summarized as below,

Combining the deep learning with IoT is proposed to perform known as well as unknown device type categorization,

Anomaly identification is enabled to protect internet of things networks from unauthorized device entry.

To reduce the dimensionality of datasets, the proposed method uses an automatic building extraction approach, which achieves equilibrium among overhead and classification accuracy.

Remaining manuscript is designed as. Literature review is presented in Section 2, Proposed Methodology is illustrated in Section 3, the results and the discussions are presented in Section 4 and Finally, Section 5 concluding this manuscript.

2. Literature review

Several works were already suggested in the literature related to IoT device type identification; a few works are divulged here,

Yousefnezhad et al., [17] have suggested Device type identification (DTI) with the use of full packet information utilizing real-time network traffic. That was the consolidation of sensor measurement and statistical feature sets. The suggested method was assessed through normal and under-attack conditions through the collection of IoT devices real-time data. It provided lower detection accuracy and higher accuracy Yue et al., [19] have presented Device type Identification using security behaviour analysis dependent on deep learning in IoT. LGBM algorithm produced excellent detection results, and handled situations for devices, that provided high accuracy and low computational time Kumar et al., [8] have presented IoT device detection through network traffic analysis. Where, network statistics and machine learning was employed for the recognition of IOT devices. Then, a differentiation between traffic produced by IOT devices and non-IOT devices using supervised training also included. It provided high accuracy and low computational time. Li et al., [9] have presented Deep learning in IOT security. It covers how IOT recognize and response to cyber assaults, to search the prospect of deep learning for enhancing the IOT security structure. It discussed about security research in application areas, like industrial IOT, the internet of vehicles, smart grid, smart homes, smart medical. Afterwards, a summary of the areas that were developed in future technological development was given, like computing power sharing via the edge network processing unit central device, integrating environmental simulation models with real world, malicious code detection, intrusion detection, production safety, vulnerability detection, fault diagnosis, and blockchain technology. It provides higher accuracy and low AUC.

Saba et al., [13] have presented Anomaly-base intrusion identification scheme for IOT utilizing deep learning. CNN-based method was presented for anomaly-based IDS that take benefit of IOT’s power to proficiently scrutinize entire traffic across the IOT. The method was trained with tested under NID, BoT-IoT data sets and reached 99.51%, 92.85%, accuracy. It provides higher AUC and high computation time.

Karthika and Vidhya Saraswathi, [6] have presented IoT under machine learning safety improvement in video steganography allocation for Raspberry Pi. The AI Security System for Raspberry Pi IoT setup uses a progression mechanism that is driven by representations. The AI Security System for Raspberry Pi IoT setup uses a progression mechanism that is driven by representations. A lower control, commercially sensible, typical IOT-dependent safety AI supports nearness ID, unmistaken 98% proof, and pariahs confirmation. The process employs USB Webcam as image capture tool and electric door strike as actuator; such devices offer application programming interfaces for gathering game plans were suited with IOT fundamental development for Raspberry Pi video steganography dissemination. It provides low computation and low accuracy.

Azumah et al., [1] have presented a deep lstm based method for intrusion identification IOT devices network in smart home to predict cyber-attacks on smart home IOT network devices. The presented method was based upon long-term memory, which accomplishes a substantial accuracy. It provides high accuracy and high computation time.

3. Proposed methodology

IoT device type identification using Training deep quantum neural networks optimized with a Chimp optimization algorithm for enhancing IOT security (IOT-DTI-TDQNN-COA-ES) is proposed here. [3]

The proposed method consists of three phases: data collection, feature extraction and detection [15]. For the collection of data, real internet traffic dataset from different IoT device types are collected. For feature extraction phase, the internet traffic features, are extracted by automated building extraction (ABE) method. IoT device type detection phase, Training deep quantum neural networks (TDQNN) [4] optimized with Chimp optimization algorithm (COA [7]. Figure 1 shows that the Block Diagram proposed IOT-DTI-AGNN-WMA approach.

Fig. 1.

Block diagram proposed IOT-DTI-AGNN-WMA approach.

3.1. Data collection

In Data collection phase, the file has collected the data of network traffic. Its setup is typically “*.pcap” and stage is commonly executed by Wire shark. In this phase, real network traffic dataset collected from different IoT device types are included. IoT device types are provided in Table 1.

Table 1
IoT device categories

Device category Model Instances

Training Testing

Lights Samsung 243 53

Smoke Detector Nest Nest_Protect 269 31

Baby monitor BeseyeBaby_Monitor,_Pro 197 103

Safety camera Simple_Home XCS7_1001 200 100 Withings WBP02_WT9510 166 134

Motion sensor D_Link DCH_S150 229 71

Watch LG G_Watch_R 200 100 Urban, Sony _SWR50 213 87

Television Samsung UA40H6300AR 200 100 UA55J5500AKXXS 251 49

Device category	Model	Instances
Lights	Samsung	243	53
Smoke Detector	Nest Nest_Protect	269	31
Baby monitor	BeseyeBaby_Monitor,_Pro	197	103
Safety camera	Simple_Home XCS7_1001 200 100 Withings WBP02_WT9510	166	134
Motion sensor	D_Link DCH_S150	229	71
Watch	LG G_Watch_R 200 100 Urban, Sony _SWR50	213	87
Television	Samsung UA40H6300AR 200 100 UA55J5500AKXXS	251	49

3.2. Automated building extraction (ABE) based feature extraction

Automatic building extraction from data plays a critical role in development in urban places and making digital city construction applications. So extraction by using automated building is extremely a challenging. The features from the internet traffic dataset are extracted using automatic building extraction. Here, the network traffic feature vector is extracted through feature extraction. The features that were used in the IoT device identification contain packet sizes; mean, variance and kurtosis are extracted using ABE method. Where $y_{1}$ is denoted as the input of the $C^{t h}$ layer, data is located at $f g$ . The result of this net work at the time step is U. Where $P_{f g h}^{e} (u)$ , is expressed in equation (1). $\begin{matrix} (1) & P_{f g h}^{e} (u) = {(W_{h}^{g})}^{U} \times y_{1}^{g (f g)} (u) + {(W_{h}^{s})}^{U} \times Y_{1}^{s (f g)} (U - 1) + C_{h} \end{matrix}$

Where $Y_{1}^{s (f g)} (U - 1)$ , $y_{1}^{g (f g)} (u)$ are the inputs of the standard convolution layer. The values of ${(W_{h}^{g})}^{U}$ and ${(W_{h}^{s})}^{U}$ are the weights of the standard convolution layer. Mean value is defined as the average mean value on region of interest and it is expressed in equation (2) $\begin{matrix} (2) & R_{e} = K_{sq} (ν_{e}) = \frac{1}{I \times W} \sum_{j = 1}^{I} \sum_{i = 1}^{W} ν_{e} (j, i) \end{matrix}$ where K is the recurrent operation, ν is denoted as output. The variance is the measurement of distance among the features extracted and it is shown in equation (3), $\begin{matrix} (3) & V = K_{ex} (r X) = ρ (r, X) = ρ (X_{2} ρ (X, r)) \end{matrix}$ where r is generated by shrinking ρ through spatial dimensions. Kurtosis is defined as the fourth power of the variance of the features among two different variables functions and it is expressed in equation (4), $\begin{matrix} (4) & \bar{W} = K_{kurtosis} (T_{e}, V_{e}) \end{matrix}$ where $K_{kurtosis}$ is recurrent operation of kurtosis, $T_{e}$ is the time variation of kurtosis and $V_{e}$ is the size of the data. However, due to the measure of cross-entropy loss on a validation set is constantly a good indicator of the quality of data. The proposed automatic building extraction method is agood choice nowadays for the network traffic feature extraction task. By using training deep quantum neural networks; the extracted features are given to IoT device type identification.

3.3. IoT device type recognition using training deep quantum neural networks

In this section, IoT device identification is done using Training Deep Quantum Neural Networks (TDQNN).The extracted feature data’s are given to the IoT device type identification for the identification of known and unknown device categories. In quantum computing principles into the neural network architecture, the proposed approach has the potential to strength quantum advantages. Quantum computing offers the possibility of exponential speedup for certain computations, which could provide significant performance gains in processing and analyzing IoT device data. TDQNN combine the power of deep learning and quantum computing, allowing for more sophisticated and nuanced analysis of IoT device data. This can result in more precise classification and identification of device types. The training Deep Quantum Neural Networks to recognize the internet of things devices plugged into the network. At this way, this is feasible to find distinct configurations that can differentiate between different device kinds. The proposed technique can protect the IoT networks’ process against various attacks effectually by analysing IoT devices’ operations.

The network architecture makes it possible to use fewer coherent qubits to hold the intermediate states necessary for CNN evaluation. Thus, the number of qubits we need to hold only scales with the network width. Quantum computers use two basis quantum states denoted as $| 0 ⟩$ and $| 1 ⟩$ .

Matric K is selected, so the cost function is raised instantly: the change in C is given in equation (5) $\begin{matrix} (5) & Δ E = \frac{C}{O} \sum_{w = 1}^{O} \sum_{m = 1}^{M + 1} u z (ρ_{w}^{1} Δ ε^{1} (σ_{w}^{1 - 1})) \end{matrix}$

From Equation (4), a formula is derived for parameter matrices where, $M + 1 = out ρ_{w}^{1}$ , $ρ_{w}^{1} = ε^{1} (\dots ε^{2} (ε^{1} (ρ_{x}^{1 - 1})) σ_{w}^{1} = F^{l + 1} (F^{M}) (F^{out})$ . The parameter matrices which is equated in (6) $\begin{matrix} (6) & F_{g}^{m} = η \frac{2^{l} m - 1}{O} \sum_{w = 1}^{O} {tr}_{train} L_{g}^{m} \end{matrix}$ where the trace is over all qubits that are not affected by $L_{g}^{m}$ and η is the learning rate. To express their basis component of information qubit is used. Qubit $| μ ⟩$ . is expressed in equation (7) $\begin{matrix} (7) & | μ ⟩ = δ | 0 ⟩ + γ | 1 ⟩ \end{matrix}$ where δ and γ are complex number, Quantum computing utilizes this property to achieve higher computational efficiency. Quantum states are denoted in equation (8) $\begin{matrix} (8) & | μ^{out}) = I | μ^{in}) \end{matrix}$ where I is an unknown operator which is used in TDQNN using $| μ_{i}^{out}$ ), $| μ_{i}^{in})$ for $i = 1, \dots, I$ . In the TDQNN contains an input layer, an output layer, and hidden layers. Similar to a classical deep quantum network, estimate the training pairs by minimizing certain cost function. The cost function is denoted in equation (9) $\begin{matrix} (9) & F = \frac{1}{M} \sum_{o = 1}^{O} (μ_{o}^{out} | ω_{E . o}^{out} | μ_{o}^{out} \end{matrix}$ where F is denoted as cost function, M is denoted as maximum value. The second property of the TDQNN is evaluated. After data processing, hey updated and get the final result that is utilized for the identification of known and unknown device categories in the newly formed graph. To get more accurate values the results are given for classification in Chimp Optimization Algorithm (COA).

3.4. Optimized TDQNN for IoT device type identification using Chimp Optimization Algorithm (COA)

COA is used for optimizing the parameters of the TDQNN model to receive the optimum parameters. These parameters are optimized to compute the optimal parameters to ensure exact categorization. The Chimp Optimization Algorithm improves the security of the proposed approach by optimizing the TDQNN for accurate device type identification. This enables tailored security measures based on device types, enhancing protection against unauthorized access. The algorithm enhances anomaly detection capabilities, aiding in the identification of security breaches and compromised devices. It facilitates rapid response to threats through fast convergence and training, mitigating potential risks. The Chimp algorithm enables fine-grained security policies by capitalizing the optimized TDQNN, addressing specific vulnerabilities of different device types for enhanced overall IoT security.

COA is developed to solve these two issues: slow convergence speed and trapping in local optimization. The weight parameters of the TDQNN $μ^{in}$ and δ are optimized using the Chimp optimization algorithm (COA).These are verified, and objective function indicates the parameter values optimization and TDQNN classifier which is mentioned in equation (10), $\begin{matrix} (10) & Fitness Function = optimizing [μ^{in}] \end{matrix}$

A fitness function is a specific type of objective function, used to summarise a given design solution to attain the goals. Here $μ^{in}$ is defined as the optimized value stored part which is mentioned in equation (11) $\begin{matrix} (11) & e = | e . w_{opt} (u) - l . w_{chimp} 9 (u) | \end{matrix}$ where u indicates the number of current iteration and $w_{opt}$ implies vector of prey position, $w_{chimp}$ implies position vector of a chimp shown in equation (12) $\begin{matrix} (12) & w_{chimp} (u + 1) = w_{opt} (u) - a . d \end{matrix}$

Chimp optimization algorithm is used to optimize weight parameter and of TDQNN classifier which accurately detects the IOT device type. At the end, TDQNN classifier accurately detects the IOT device type using COA. The Chimp Optimization Algorithm optimizes the DQNN’s parameters, making it more robust to variations and noise in the input data. This robustness can help mitigate the challenges posed by the heterogeneity and dynamic nature of IoT environments. The optimized DQNN can better handle variations in device behavior, environmental conditions, and data quality. The COA enhances the agility of the TDQNN by enabling faster convergence and training. This adaptability allows the system to respond more efficiently to emerging security threats and changes in the IoT landscape. It can quickly adapt to new device types, patterns, or attack vectors, improving the system’s ability to detect and counter potential security risks. Accurate device type identification enables more effective security measures. The proposed approach can enforce tailored security policies based on identified device types. This enables finer-grained access control, anomaly detection, and security protocols, enhancing overall IoT security.

4. Results and discussion

This segment describes the IoT device type identification using Training Deep Quantum Neural Networks with Chimp Optimization Algorithm for Enhancing IoT Security (IoT-DTI-TDQNN-COA-ES). The simulations are done in Python, Mac-OS Catalina operating system. The performance metrics is evaluated, like ROC, computational time and accuracy and the proposed method is comparing to existing IoT-DTI-STI-ES and IoT-DTI-DNN-ES models.

The real network traffic datasets gathered from ten different kinds of Internet of Things (IoT) devices, including baby monitors, lights, motion detectors, security cameras, smoke detectors, sockets, thermostats, TVs, watches, and water sensors.. Some kinds are made by various vendors, or by the same sellers but with various models. In a pcap file, traffic data passing via a switch connecting all of the devices is first recorded using Wireshark, and recording data has transformed as 297-feature vectors. A count of destination port, size of packet, TCP segments count out of order, are some examples of such features. TCP segment represents an effectual connection that begins with 3-way-handshake and discontinues when times out. The training data set contains 1400 feature vectors collected via 7 internet of things devices and labelled its device-type. The testing data set has 1005 feature vectors gathered via 10 device categories along 3 (out of 10) are presented devoid of label, so this is deemed as ‘anomaly’. That means 3 of 10 device types are deemed as newly/unknown/unseen, 7 are deemed as known/seen. Table 2 shows that the Output of IoT Device.

Table 2
Output of IoT Device

Device category Model # of instances

Training Testing

Baby monitor Beseye Baby_Monitor,_Pro 200 100

Lights Samsung 200 100

Motion sensor D_Link DCH_S150 200 100

Security camera Simple_Home XCS7_1001 Withings WBP02_WT9510 200 100

Smoke identifier Nest Nest_Secure 200 100

Television Samsung UA40H6300AR UA55J5500AKXXS 200 100

Watch LG G_Watch_R Urban,Sony _SWR50 200 100

Unknown1 N/A 0 100

Unknown2 N/A 0 100

Unknown3 N/A 0 100

Device category	Model	# of instances
Baby monitor	Beseye Baby_Monitor,_Pro	200	100
Lights	Samsung	200	100
Motion sensor	D_Link DCH_S150	200	100
Security camera	Simple_Home XCS7_1001 Withings WBP02_WT9510	200	100
Smoke identifier	Nest Nest_Secure	200	100
Television	Samsung UA40H6300AR UA55J5500AKXXS	200	100
Watch	LG G_Watch_R Urban,Sony _SWR50	200	100
Unknown1	N/A	0	100
Unknown2	N/A	0	100
Unknown3	N/A	0	100

Experimental setup. The proposed technique is activated in Python3 on MacOS Catalina OS. To conduct min-max scaling of data, built-in MinMaxScaler for Python Sklearn package is utilized because distinct features have different value ranges in the raw data set, causing all feature values to fall in $[0, 1]$ range. The experiments are used to train and categorizing known devices. The classifier utilizes training Deep Quantum Neural Networks through Sklearn then set its tool-kit parameters: n_estimators = 200, a functioning to scale the superiority of splitting is criteria = entropy, max_depth = 15, min_samples = 5. Chimp is considered as an optimization approach, including $10^{- 3}$ constant learning rate. Because AE can display stable losses after 30 epochs, it is trained for 100 epochs. The parameters employed in OPTICS, minP t = 50 is set when additional parameters set to defaulting values offered through Sk learn library.

4.1. Performance metrics

Analysis of performance metrics such as accuracy, computational time and ROC are done.

4.1.1. Accuracy

The ratio of exact predictions to the total number of proceedings is determined in equation (13), $\begin{matrix} (13) & Accuracy = \frac{W_{Q} + W_{P}}{W_{S} + W_{P} + J_{Q} + J_{T}} \end{matrix}$ where ( $w_{Q}$ ) is known as True Positive, ( $W_{P}$ )is known as True Negative, ( $J_{Q}$ ) is known as False Positive and ( $J_{R}$ ) is known as False Negative

4.1.2. ROC

Roc gives insight and its clearance rate from the body is expressed in equation (14) $\begin{matrix} (14) & ROC = 0.5 \times (\frac{W_{Q}}{W_{Q} + J_{R}} + \frac{W_{P}}{W_{P} + W_{Q}}) \end{matrix}$

4.1.3. Computation time

Computation time is an operating time, where the duration of time needed to act a computational process is called as computation time. It is shown in equation (15). $\begin{matrix} (15) & Computation time = \frac{W_{c} * WCF}{Clock Rate} \end{matrix}$ where $W_{c}$ is denoted as clock cycles, $WCF$ is said as instruction count in clock cycle.

4.2. Simulation result

Figures 2–4 shows the simulation results of the IoT device type identification using Training Deep Quantum Neural Networks with Chimp Optimization Algorithm for Enhancing IoT Security (IoT-DTI-TDQNN-COA-ES). The performance metrics like accuracy, ROC and time calculation are analysed and compared with the existing IoT-DTI-STI-ES and IoT-DTI-DNN-ES methods.

Fig. 2.

Comparative analysis of accuracy with various methods.

Figure 2 shows Accuracy analysis. Here, the proposed method provides 5.88% and 15.38% higher accuracy for known IoT devices; 21.78% and 8.22% higher accuracy for unknown IoT devices compared with existing methods like IoT-DTI-STI-ES and IoT-DTI-DNN-ES respectively.

Figure 3 represents ROC analysis. The proposed method attains 35.66% and 28.89% with existing IoT-DTI-STI-ES and IoT-DTI-DNN-ES methods.

Fig. 3.

Comparative analysis of ROC with various methods.

Figure 4 depicts the analysis of computational time. Here, the proposed method attains 28% and 31% Low computational time compared with existing methods.

Fig. 4.

Comparative analysis of computational time.

Discussion.

Table 3

Bench mark table using literature support

Methods	Performance analysis

	Accuracy (%)	ROC	Computation time (s)
Yousefnezhad et al., [17]	78.11	0.85	180
Yue et al., [19]	81.25	0.91	220
Kumar et al., [8]	85.18	-	367
Li et al., [9]	-	0.81	-
Saba et al., [13]	90.56	0.62	-
Karthika and Vidhya Saraswathi, [6]	96.74	-	248
Azumah et al., [1]	-	0.07	193
Balakrishnan et al., [2]	95.65	-	-
Yin et al., [16]	98.66	-	198
Sharma et al., [14]	85.45	0.56	-
IOT-DTI-TDQNN-COA-ES (proposed)	99.14	0.97	70

In this manuscript, training Deep Quantum Neural Networks Optimized with Chimp Optimization Algorithm is an interesting approach to enhance Iot security and IoT Device Type Identification. It is clear to see that the COA-TDQNN algorithm has advantages over TDQNN in every way. First off, the COA-TDQNN algorithm’s cross-entropy loss value is much lower than that of the TDQNN method throughout every training cycle on training set, but the accurateness is greater. This shows that the COA-TDQNN has a greater training impact, demonstrating its effectiveness in model training. Additionally, the pattern of discount is evident from the perspective of the validation set. With greater precision, the model’s total detection rate increases. As a result, the effectiveness of the model’s prediction is confirmed by the proposed algorithm. According to the practical significance, a method with an entire higher identification rate of network attacks is essential because the small value of cross-entropy loss function determines the high accuracy. The research demonstrates the successful application of TDQNN for identifying the types of IoT devices with respect to their data patterns, contributing to enhancing IoT security. Address the privacy and ethical considerations associated with IoT device type identification. Investigate techniques such as machine learning or differential privacy to protect sensitive data and preserve user privacy. By optimizing the TDQNN using the Chimp optimization algorithm, the accuracy of device type identification is improved, leading to more reliable and robust security measures. Simulation results illustrate from Table 3 shows that the Proposed IOT-DTI-TDQNN-COA-ES model provide 4.52%, 13.82%, 9.45%, 11.45%, 12.45%, 15.34% and 11.94% higher accuracy compared with existing methods like, Yousefnezhad et al., [17], Kumar et al., [8], Saba et al., [13], Karthika and Vidhya Saraswathi, [6], Balakrishnan et al., [2], Yin et al., [16] and Sharma et al., [14] respectively. The Proposed IOT-DTI-TDQNN-COA-ES model provide 22.33%, 28.70%, 14.56%, 11.22%, 22.22%, and 15.21% higher ROC compared with existing methods like Yousefnezhad et al., [17], Yue et al., [19], Li et al., [9], Saba et al., [13], Azumah et al., [1] and Sharma et al., [14] respectively. The Proposed MLBF-ArOA-SRCE model provide 10.22%, 24.67%, 15.7%, 32.22%, and 18.11% lower computation time compared with existing methods likes, Yousefnezhad et al., [17], Kumar et al., [8], Karthika and Vidhya Saraswathi, [6] Azumah et al., [1] and Yin et al., [16] respectively.

5. Conclusion

In this manuscript, IoT device type identification using Training Deep Quantum Neural Networks with Chimp Optimization Algorithm for Enhancing IoT Security (IoT-DTI-TDQNN-COA-ES) is verified. Here, the assessment metrics like accuracy, computational time and ROC are analyzed. Then the proposed IoT-DTI-TDQNN-COA-ES technique shows lower computational time at 28.21% and 28.99%compared with the existing IoT-DTI-STI-ES and IoT-DTI-DNN-ES methods.

The research demonstrated the effectiveness of TDQNN for recognizing the types of IoT devices based upon their data patterns, contributing to enhancing IoT security. By optimizing the TDQNN using the Chimp optimization algorithm, the accuracy of device type identification was improved, resulting in more reliable and robust security measures. The combination of TDQNN and the Chimp optimization algorithm showcased the potential for advanced machine learning techniques and optimization algorithms to strengthen IoT security measures. The research focused on device type identification, and further exploration of other security aspects, such as anomaly detection, intrusion prevention, and authentication, would enrich the overall IoT security framework. Future work will focus on expanding the variety and quantity of internet of things devices and collecting real-time network data for long period. Extend the research to include additional security aspects beyond device type identification, such as anomaly detection, real-time threat response, and secure communication protocols. Collaborate with IoT industry stakeholders and security experts to validate and deploy the proposed approach in real-world IoT ecosystems, considering practical challenges and implementation considerations. Also decide to include behavioral features for device recognition.

Conflict of interest

The authors have no conflict of interest to report.

References

S.W.

Azumah,

Elsayed,

Adewopo,

Z.S.

Zaghloul and

Li, A deep lstm based approach for intrusion detection iot devices network in smart home, in: 2021 IEEE 7th World Forum on Internet of Things (WF-IoT), IEEE, 2021 pp. 836–841.

Balakrishnan,

Rajendran,

Pelusi and

Ponnusamy, Deep belief network enhanced intrusion detection system to prevent security breach in the Internet of things, Internet of things 14 (2021), 100112. doi:10.1016/j.iot.2019.100112.

Bao,

Hamdaoui and

W.K.

Wong, Iot device type identification using hybrid deep learning approach for increased IoT security, in: 2020 International Wireless Communications and Mobile Computing (IWCMC), IEEE, 2020 June, pp. 565–570.

Beer,

Bondarenko,

Farrelly,

T.J.

Osborne,

Salzmann,

Scheiermann and

Wolf, Training deep quantum neural networks, Nature communications 11(1) (2020), 808. doi:10.1038/s41467-020-14454-2.

Dev,

S.A.

Khowaja,

P.K.

Sharma,

B.S.

Chowdhry,

Tanwar and

Fortino, DDI: A novel architecture for joint active user detection and IoT device identification in grantfree NOMA systems for 6G and beyond networks, IEEE Internet of Things Journal 9(4) (2021), 2906–2917. doi:10.1109/JIOT.2021.3095255.

Karthika and

Vidhya Saraswathi, IoT using machine learning security enhancement in video steganography allocation for Raspberry Pi, Journal of Ambient Intelligence and Humanized Computing 12 (2021), 5835–5844. doi:10.1007/s12652-020-02126-4.

Khishe and

M.R.

Mosavi, Chimp optimization algorithm, Expert systems with applications 149 (2020), 113338. doi:10.1016/j.eswa.2020.113338.

K.R.

Kumar,

Hemanth,

C.A.

Kumar,

K.M.

Sahith and

G.A.

Prasanth, IoT device identification through network traffic analysis, Int. Res. J. Modern. Eng. Technol. Sci 2(06) (2020).

Li,

Zuo,

Song and

Lv, Deep learning in security of Internet of things, IEEE Internet of Things Journal 9(22) (2021), 22133–22146. doi:10.1109/JIOT.2021.3106898.

10.

Luo,

Chen,

Ge,

Feng and

Lu, Transformer-based device type identification in heterogeneous IoT traffic, IEEE Internet of Things Journal (2022).

11.

Meidan,

Bohadana,

Shabtai,

Ochoa,

N.O.

Tippenhauer,

J.D.

Guarnizo and

Elovici, Detection of unauthorized IoT devices using machine learning techniques, 2017, arXiv preprint arXiv:1709.04647.

12.

Qing,

Wang,

Guo and

Yang, Device type identification via network traffic and lightweight convolutional neural network for Internet of things, IEEE Access 8 (2020), 200219–200228. doi:10.1109/ACCESS.2020.3032469.

13.

Saba,

Rehman,

Sadad,

Kolivand and

S.A.

Bahaj, Anomaly-based intrusion detection system for IoT networks through deep learning model, Computers and Electrical Engineering 99 (2022), 107810. doi:10.1016/j.compeleceng.2022.107810.

14.

Sharma,

Namasudra,

R.G.

Crespo,

Parra-Fuente and

M.C.

Trivedi, EHDHE: Enhancing security of healthcare documents in IOT-enabled digital healthcare ecosystems using blockchain, Information Sciences 629 (2023), 703–718. doi:10.1016/j.ins.2023.01.148.

15.

Tang,

Ma,

Cui,

Ji and

Abraham, Long text feature extraction network with data augmentation, Applied Intelligence (2022), 1–16.

16.

Yin,

Yang,

Wang and

Dai, Iot etei: End-to-end iot device identification method, in: 2021 IEEE on Dependable and Secure Computing (DSC), IEEE, 2021, pp. 1–8.

17.

Yousefnezhad,

Malhi and

Främling, Automated iot device identification based on full packet information using real-time network traffic, Sensors 21(8) (2021), 2660. doi:10.3390/s21082660.

18.

Yu,

Zhang,

Chen,

Ma and

Chen, Large-scale IoT devices firmware identification based on weak password, IEEE Access 8 (2020), 7981–7992. doi:10.1109/ACCESS.2020.2964646.

19.

Yue,

Li,

Legg and

Li, Deep learning-based security behaviour analysis in IoT environments: A survey, Security and Communication Networks 2021 (2021), 1–13.