An optimized deep learning model to ensure data integrity and security in IoT based e-commerce block chain application

Abstract

E-commerce, often known as electronic commerce, is the purchasing and selling of goods over the internet using electronic devices to share data. Banks and other financial institutions are frequently added as third-party platforms to traditional e-commerce platforms. As a result, it raises issues with integrity and cyber security. We suggest a deep learning-based strategy called the Hybrid Interactive Autodidactic School-Based Teaching-Learning Optimization (HIASTLO) algorithm to address these issues. The IoT-based e-commerce blockchain is used to extract and reject the various cyberattacks in the network, and deep learning is utilised to improve the weight and bias of the neural networks. We used a variety of performance indicators, including accuracy, precision, and recall, to identify cyberattacks. We also evaluated how well our work performed when compared to previous BSIoTNET, BCFC, DRNN, DNN-KNN, MOO-FS, LRNN, and HDLM efforts. Furthermore, MudraChain and NormaChain are used to examine the transaction time of our suggested task. The results show that our suggested work performs better than any other methods and offers highly secure internet services.

Keywords

E-commerce blockchain deep learning cyber attacks IoT

1 Introduction

The global business community has experienced an immediate expansion of electronic commerce as a result of the advent of globalisation in economics and communication technology [1]. E-commerce is the practise of conducting business using electronic media and the internet. As a result, traditional business practises have been fundamentally altered. Additionally, it reduces pollutants, which has a big effect on the environment. E-commerce is typically conducted through a website gateway, which makes use of the electronic shopping cart system and allows customers to pay with credit cards, debit cards, or EFT (Electronic Fund Transfer) systems [2]. Trust between suppliers and customers are crucial and extremely significant in the world of e-commerce.

With the advent of fifth-generation (5 G) technology, numerous IoT applications have recently been developed to improve the Quality of Service (QoS) and user experience, including smart transportation, healthcare, and virtual and augmented reality experiences. Reduced latency, increased system capacity, high data rates, and energy efficiency are just a few of the distinctive qualities that the 5G-enabled IoT revolution enables. However, such a revolution also results in a significant increase in data generation, which further increases the need for sophisticated and efficient network-wide data analytic operations. Data security and privacy issues, like as breaches and the loss of sensitive data, are also exacerbated by data growth. The requirements of a 5G-enabled IoT, particularly its distinct characteristics of low latency and high throughput, are not satisfied by standard data analysis and security methods.

Recently, IoT-based e-commerce platforms with a collection of dispersed, tiny IoT devices have been formed. These dispersed devices need to be managed by a structure. Thanks to blockchain technology, it is now possible. Nothing more than distributed data made up of several transaction records can be called blockchain. These records are linked together by several chains [3]. By organising the IoT’s distributed devices and fostering trust, blockchain solutions use more secure cryptographic methods. Utilizing the hash-connected chain data structure of the blockchain also allows for complete data integrity. By restricting access to just authorised users, it also provides high-level confidentiality protection for transactional information. The authorization is recognised by providing a time-sensitive unique key that immediately disables the system when the duration runs out. However, blockchain technology does not need a central authority because it uses decentralised computation [4].

In order to oversee the security of the user’s transaction information, some regulators are also required to develop the decryption procedure. It can protect transaction data and thwart unauthorised attacks. Some systems also work around this problem by storing the data in plaintext, although doing so compromises user privacy and the security that was promised. It is essential to keep track of transaction records and to protect those using encrypted keys. Therefore, a deep learning approach called the Hybrid Interactive Autodidactic School-Based Teaching and Learning Optimization (HIASTLO) algorithm was proposed.

The significance of our work is described here.

In IoT-based e-commerce applications, our suggested blockchain is utilised to detect cyberattacks such self-serving mining attacks, DDoS attacks, poisoning attacks, Sybil attacks, and impersonation attacks.

It protects the network’s IoT-based e-data commerce’s integrity.

Cyberattacks can be quantified using performance indicators including recall, accuracy, and precision.

The deep learning neural network’s weight and bias are improved with the help of the HIASTLO optimization algorithm.

The remainder of the work is organized as Section 2 deliberates the background of our work. Section 3 elaborates on the deep learning optimization model of our work. The proposed work is explained in Section 4 and the simulation results and the comparative analysis are listed in section 5. Finally, we concluded the proposed work in Section 6.

2 Related works

BlockSecIoTNet, a decentralised security architecture based on Software Defined Networking (SDN), was introduced by Rathore et al. [5] for IoT networks in the smart city. Three key strategies efficiently identify IoT network assaults. They are SDN, Blockchain, and Fog and mobile edge computing. Compared to other methods already in use, this method achieved faster detection times with higher accuracy. Malomo et al. [6] proposed the idea of a blockchain-enabled federated cloud to identify cybersecurity risks. Through the use of the Dempster-Shafer theory, the breach detection gap (BDG) is reduced, necessitating constant monitoring and analysis of network data. Last but not least, the federated cloud framework powered by blockchain surpassed traditional methods with greater results..

Almiani et al. [7] presented the Deep Recurrent Neural Network (DRNN) for IoT intrusion detection systems. Multi-layered recurrent neural networks are created for Fog computing security. Cohen’s Kappa coefficients and Mathew correlation measures are used to assess the effectiveness of DRNN in identifying attacks. This effectively proves the stability and resilience of simulation. Recursive neural networks are effectively used to tweak the hyperparameters in order to improve intrusion detection. The investigations into the U2 R, R2 L, and Probe attacks are completed. De Souza et al. [8] presented the Deep Neural Networks (DNN) and k-Nearest Neighbor (kNN) technique for intrusion detection in fog-based IoT environments. The CICIDS2017 and NSL-KDD databases do the experimental validations. The accuracy of the NSL-KDD and CICIDS2017 datasets was 99.77 percent and 99.85 percent, respectively. Finally, the reduced overhead is achieved with the least amount of processing and memory cost, but it requires a longer execution time.

Verma et al. [9] suggested machine learning classification algorithms to protect the IoT from DoS assaults. Utilizing the NSL-KDD UNSWNB15 and CIDDS-001 datasets with prominent metrics, the classifier’s performance is evaluated. In terms of response time, Raspberry Pi is used to validate the IoT-specific hardware classifiers. Extreme gradient boosting and trees outperformed superior response time and important metric trade-offs. For detecting routing attacks in IoT networks, this machine-learning categorization is not appropriate. Roopak et al. [10] presented the multi-objective optimization-based feature selection (FS) model (MOO-FS) to identify DoS attacks in IoT. With the help of the feature selection model, the performance of intrusion detection is enhanced while dimensionality is decreased. The optimization model successfully improves the data that may be deceptive in terms of attack detection, such as accuracy and relevance. The MOO-FS had a 99.9% accuracy rate when detecting DDoS attacks.

Latif et al. [11] introduced the Lightweight Random Neural Network (LRNN) for Industrial Internet of Things attack detection (IIoT). The LRNN approach foresees a variety of cybersecurity assaults, including as denial of service, malicious control, scanning, surveillance, and data type probing (DoS). In addition, LRNN outperforms other approaches including decision trees, support vector machines, and conventional artificial neural networks in terms of F1 score, recall, precision, and accuracy for intrusion detection. A Hybrid deep learning model for Botnet Attack Detection in IoT was proposed by Popoola et al. [12]. The Long Short-Term Memory Autoencoder (LAE) reduces the dimensionality of the large-scale IoT network traffic data during the encoding step. Zhou et al. (2017) [20] proposed the efficiency of resource sharing and optimal allocation in CMfg systems that are impacted by the service composition with the best overall quality of service (QoS). However, there are still certain restrictions on how to compose manufacturing services, particularly in the complex cloud environment. Dorri et al. (2017) [21] suggested a minimal BC-based architecture for the Internet of Things, which mostly does away with the overheads of traditional BC while retaining the majority of its security and privacy benefits. In order to reduce energy usage, IoT devices benefit from a private, irreversible ledger that functions similarly to BC but is centrally controlled. Shafay et al. (2022) [22] examined the significance of combining deep learning and blockchain technologies. We examine the body of research on deep learning and blockchain integration. Rathor et al. (2021) [23] suggested a Deep Learning (DL) and blockchain-enabled security architecture for intelligent, 5G-enabled IoT that makes use of DL capabilities for operations involving intelligent data analysis and blockchain for data security. Intelligent cities seek efficient and long-lasting smart solutions for government, energy, transportation, and environment issues due to the rapid growth in population. The smart city platform, which combines IoT, Big Data, and the Internet of Energy, is one of the most practical alternatives. It has a number of problems, including insufficient IoT security, difficulties maintaining and improving efficiency, higher operating costs associated with the construction of large data centres, good permeability to damage, difficulty gaining the trust of electricity internet users, rapid leakage of consumer privacy, an unacceptable business model, etc. One of today’s most revolutionary technologies is blockchain. Numerous towns all over the world are initiating blockchain initiatives as part of the wider endeavour to build the urban future. Digital transformation has numerous potential benefits, but it also raises important concerns including data security and confidentiality. This study suggested a security architecture that makes use of smart devices and blockchain technology to offer a secure communication system in an intelligent city [24 –26].

The existing methodologies are summarized in Table 1.

Table 1
Summary of existing methodologies

Author Method Name of the datasets Merits Demerits

Rathore et al. [5] BlockSecIoTNet NSL-KDD dataset Takes lower time to alleviate attacks Higher latency and cost of computation

Malomo et al. [6] Blockchain-enabled federated cloud framework NSL-KDD dataset Effectively minimize the BDG Storage constraints

Almiani et al. [7] Deep Recurrent Neural Network NSL-KDD dataset Investigates U2 R, R2 L, and Probe attacks Higher computational overhead

De Souza et al. [8] DNN-kNN CICIDS2017 and NSL-KDD dataset Minimum processing and memory cost Takes more execution time

Verma et al. [9] Machine learning classification algorithms NSL-KDD, UNSWNB15, and CIDDS-001 datasets The excellent trade-off among response time and prominent metrics Not suitable for routing attacks detection in IoT networks

Roopak et al. [10] MOO-FS CICIDS2017 dataset Optimize the accuracy and irrelevance of data Computationally expensive

Latif et al. [11] Lightweight Random Neural Network (LRNN) KDD-CUP 99 and NSL-KDD dataset Higher detection accuracy Higher latency

Popoola et al. [12] The hybrid deep learning model (HDLM) BoT-IoT dataset It avoids under and over fittings and required lower memory space Higher computational cost

Author	Method	Name of the datasets	Merits	Demerits
Rathore et al. [5]	BlockSecIoTNet	NSL-KDD dataset	Takes lower time to alleviate attacks	Higher latency and cost of computation
Malomo et al. [6]	Blockchain-enabled federated cloud framework	NSL-KDD dataset	Effectively minimize the BDG	Storage constraints
Almiani et al. [7]	Deep Recurrent Neural Network	NSL-KDD dataset	Investigates U2 R, R2 L, and Probe attacks	Higher computational overhead
De Souza et al. [8]	DNN-kNN	CICIDS2017 and NSL-KDD dataset	Minimum processing and memory cost	Takes more execution time
Verma et al. [9]	Machine learning classification algorithms	NSL-KDD, UNSWNB15, and CIDDS-001 datasets	The excellent trade-off among response time and prominent metrics	Not suitable for routing attacks detection in IoT networks
Roopak et al. [10]	MOO-FS	CICIDS2017 dataset	Optimize the accuracy and irrelevance of data	Computationally expensive
Latif et al. [11]	Lightweight Random Neural Network (LRNN)	KDD-CUP 99 and NSL-KDD dataset	Higher detection accuracy	Higher latency
Popoola et al. [12]	The hybrid deep learning model (HDLM)	BoT-IoT dataset	It avoids under and over fittings and required lower memory space	Higher computational cost

3 Formulation of deep learning optimization model

In this section, we formulate the deep learning optimization model, which is the combination of the Hybrid Interactive Autodidactic School-based Teaching Learning Optimization (HIASTLO) algorithm with deep learning. The steps involved in deep learning optimization techniques are delineated below:

3.1 Hybrid interactive autodidactic school-based teaching learning optimization (HIASTLO) algorithm

Brand-new swarm intelligence metaheuristic called the HIASTLO algorithm has been touted as an efficient method for resolving a variety of optimization problems. We can see that the degrees of teaching success of two teachers differ when they are teaching the same subject to students with the same academic status in two different courses. Their main difference is that pupils with good teachers earn greater average levels of merit performance. Additionally, when students interact with their peers, they learn. The core of HIASTLO is this instructional strategy. A mathematical model is developed and used for HIASTLO-based optimization. Teachers are chosen because they are the best learners because they are the persons in the population with the most knowledge. By imparting knowledge to the class, the instructor increases knowledge and helps students succeed academically. A teacher can enhance the level of his or her students by raising the class’s overall average achievement. The efficacy of a teacher is determined by the average knowledge level reached by the class’s students, which is influenced by both the learner and instructor quality. Finding a new, more competent instructor may be necessary to raise standards even higher when a teacher’s efforts to improve class quality fail. The mean value of the population is utilised to assess the calibre of the students. The teacher puts a lot of effort into raising the calibre of the students, but eventually they might require a different teacher who is more capable than they are. A new teacher will therefore start a brand-new teaching method. HIASTLO is a population-based method that uses individual user answers to identify the overall ideal, similar to other nature-inspired algorithms. Every member of the TLBO population is treated as a student in a class, and each student in the class has a fitness value. The teacher is the one who has the highest fitness value. The two phases of HIASTLO are the instructor phase and the learner phase. Based on its adaptability and diversity, the metaheuristic technique is referred to in a variety of categories. The IAS algorithm’s fundamental concepts are adopted by the community of self-learning students’ knowledge growth process. The grades of the students reflect their most advanced level of learning. Because of the gradual expansion of community collective knowledge, competition for the top spot is ongoing. One of the community members gains certain degrees of knowledge from the didactic performance. Knowledge expansion is facilitated by leadership rivalry, criticism, interactive conversation, self-learning, and teaching. Randomly initialise the student population as the beginning population in accordance with the IAS method. The student’s eligibility to enrol in IAS is determined by the upper and lower boundaries. The highest grade a student receives at any level qualifies them for the position of top student. The IAS optimization achieves the optimum performance while achieving the smallest cost function value. The technique for generating and evaluating student competency in the classroom is described in the following mathematical model.

Step 1: Randomly generate the first group of students for the school using N student, the number of design variables, and the following formula (Note that a school must have at least three students).

Step 2: Among the original students, award the best student with the best grade (in this case, the best grade equates to the lowest value of the cost function) the position of the new leading student.

Step 3: This lesson may enhance the school’s performance and the knowledge of the top student. The allowed range between the lower and upper boundaries of the design variables is where the new student is formed. The fourth student is chosen as the new school’s top student since she received the highest grade out of all the kids.

For k = 1 : M _ student

P_k = L_b + r_k (0, 1) * (U_b - L_b)

N_k = F (P_k)

End For

f (L_s) = min{ N }

Hence, the U_b and L_b are the upper and lower bound variables and the leading student is L_s. The kth student is generated as P_k in which the variables tend to the interval [o, 1]. Where, the kth student’s mark is Nk and the number of students is 1, M _ student.

3.1.1 Individual training session

The learning-directed interaction period among the leading and other students describes the individual training. Select two members from a random group of trailing students. Note the individual knowledge of the student grows in such a peer-to-peer discussion with the leader. Thereafter the individual learning session is based on their original competence that the knowledge level of the trailing students [13]. The following mathematical models formalize the individual sessions.

For k = 1 : M_student

Randomly choose P_j

Where k ≠ j

${HP}_{k}^{*} = H P_{k} + r_{k} (0, 1) * (L_{s} - I D_{k} - H P_{k})$ ;

${HP}_{j}^{*} = H P_{j} + r_{j} (1, 2) * (L_{s} - I D_{j} - H P_{j})$ ;

End For

The better marks than HP_k and HP_j is achieved then accept ${HP}_{k}^{*}$ and ${HP}_{j}^{*}$ .

Hence, the initial and second trailing students are ${HP}_{k}^{*}$ and ${HP}_{j}^{*}$ with its intrinsic competence are ID_k and ID_j, which are randomly determined as either one or two.

3.1.2 Cooperative training session

Every trailing student will contain the chance to review the final session after the individual training session. The mathematical representation of the collective session is described below:

For k = 1 : M _ student

DD_kj = (DD_k * HP_k + DD_j * HP_j)/DD_k + DD_j (DD_k + DD_j)

${HP}_{k}^{*} = H P_{k} + r_{k} (1, 2) * (L_{s} - D D_{k} - D D_{kj})$ ;

${HP}_{j}^{*} = H P_{j} + r_{j} (1, 2) * (L_{s} - D D_{j} - D D_{kj})$ ;

End For

Where, recognize ${HP}_{k}^{*}$ and ${HP}_{j}^{*}$ represent the optimal marks than HR_j and HR_k. The collective capability category is defined as DD_kj dependent on the weighted mean of the skills of students. The multi random vectors among one and two are r_k (0, 1) and r_j (1, 2).

With the use of a teacher and learning phase included in the teaching-learning-based optimization (TLBO) algorithm, also known as the Hybrid Interactive Autodidactic School-based Teaching Learning Optimization (HIASTLO) algorithm, [14] the challenges facing a new student in IAS optimization are increased.

3.1.3 Teacher phase

Here, the entire performance of the population is improved via the teacher. The random number s lies between 0 and 1. Each subject for every student Nk has calculated the mean values. The new values Y_j,k new replace the individual population Y_j,k old [14]. Equation (1) calculates the new values. $Y_{j, k new} = Y_{j, k old} + q (Y_{teacher k} - N_{k})$ (1)

Repeat the process for all populations. Calculate the fitness values of the population after updating the design variables then collect the design variables.

3.1.4 Learning phase

The student enhances themselves with the interaction between them in the learning phase. Select the two populations (u, v) randomly then compare the fitness values (f_u, f_v). $Y_{j, k new} = Y_{j, k old} + q (Y_{uk} - Y_{vj}) if f_{u} < f_{v}$ (2) $Y_{j, k new} = Y_{j, k old} + r (Y_{vk} - Y_{nk}) if f_{v} < f_{u}$ (3)

Calculate the new fitness values and modified the design variables. At last, select the best fitness value among the population for the present iteration. Repeat the process till the stopping criterion is met.

a. Deep learning with optimization (Deep Neural Network (DNN) based (HIASTLO)) algorithm:

The wonderful biological-inspired programming paradigm is neural networks. Deep learning or deep neural network (DNN) is a powerful technique for learning in neural networks. In general, the deep learning model solves several issues in the case of natural language processing, speech recognition, and image recognition [15]. Several layers perform the deep neural networks (DNN) composes computation. Normally, the DNN consists of one input layer, two hidden layers, and one output layer with neurons, weights, bias, and activation functions. Where denotes the hidden layer outputs with L hidden layers are calculated using Equation (4). $\begin{matrix} F (Y) = F [b^{(L + 1)} (g^{(L)} (b^{(L)} (. . . (g^{(2)} (b^{(2)} \\ (g^{(1)} (b^{(1)} (Y)))))))) \end{matrix}$ (4)

Where, the bias and activation of k^th neuron in the L layer is $b_{k}^{L}$ and $A_{k}^{L}$ . The activation function is described in Equation (5). $A_{k}^{L} = α (\sum_{j} W_{kj}^{L} + b_{k}^{L})$ (5)

Hence, $W_{kj}^{L}$ is the weight function of overall neurons j in the (L-1) ^th layer. The objects in which the DNN develops compositional models are expressed by the layered composition of the primitives. The additional layers enable-feature compositions from lower layers. The classic DNN has a few flaws from incorrect weight and bias optimization and complex non-linear interactions. With the aid of deep learning and the optimization (Deep Neural Network (DNN) based (HIASTLO)) techniques, certain parameters in DNN are successfully optimised. Thus, the weight and bias function of DNN are effectively optimised by the HIASTLO algorithm. Figure 1 shows the deep learning optimization model.

Fig. 1

Deep learning with an optimization model.

4 Proposed methodology

4.1 BlockChain

In the cryptocurrency world blockchain is the most important as well as a confusing term. BlockChain is exploited to support the digital currency space. It was first invented by Satoshi Nakamoto in the year 2008 and was used to act as a public transaction ledger of the cryptocurrency bitcoin. Thus it averts the dependency of the central server by ignoring the double-spending problems. Basically, it is a combination of various existing methods and it contains three basic blocks;

private-key cryptography

shared ledger

accounting and transaction details of the related network

4.2 Data pre-processing and analysis

The non-saturated data such as voice, texts, and images are collected in this stage. After that, we handle the pre-processing to extract the fitted data for further process. Convert the collected data in order to allow them in the Deep learning model in case of combining existing data properties, deleting or selecting data properties, and deleting or filling missing values. Apply the analyzed data to deep learning with optimization involving data learning using limited data, extracting inference and exploration data, mapping data, and exploring standardized data patterns. In general, a few noises with no consistency present in raw data in the data collection course, which is not appropriate for further process. But, data quality improvement with higher performance, accuracy, reliability, and quality is an important one. Hence, the data conflicts coordination, inconsistent data deletion, overlapping data elimination, and data error modification based on insight and professionalism conducts the data organization and analysis. For the development of Deep learning with optimization, enough quantity of high-quality data obtaining is complex. The proposed workflow model is delineated in Fig. 2.

Fig. 2

Proposed workflow diagram.

4.2.1 Various cyber security attacks in IoT-based E-Commerce blockchain applications

In general, sufficient ground results are unable to provide with Deep learning technology, which is constrained to the passive recognition area and the operating process is not completely transparent. In this article, we have chosen five attacks namely selfish mining attack, DDoS attack, Poisoning attack, Sybil attack, and Impersonation attack. The rapid increase in IoT nodes that have insufficient security features is often prone to different types of cybersecurity attacks. This work deals with the e-commerce industry that utilizes blockchain. Even though blockchain provides integrity to different attacks, it is prone to different cybersecurity attacks listed below:

Selfish Mining attack: In this attack, the abnormal node diverts its normal behavior without suddenly revealing the details of the newly added block. It sabotages the nodes which have high processing power by giving an unequal share of rewards.

Distributed Denial of Service(DDOS): Here a hacker loads the network with a massive amount of requests to disrupt the proper functioning of the system. Blockchain-based e-commerce is also prone to DDoS attacks. In the blockchain network, the DDOS attack enables the hacker to access the confidential network that provides the e-wallet and exchange services.

Sybil attack: In this attack, the hacker hacks the e-commerce website by using different user profiles. In the blockchain-based e-commerce network, the hacker runs multiple nodes in the network. This attack prevents the blockchain from adding new blocks into the network.

Poisoning Attack: The poisoning attack is capable of injecting malicious data into the model’s training data. In the blockchain, the poisoning attack pollutes the network by adding an increased amount of malicious data at a lower cost.

Impersonation attack: In the impersonation attack, the attacker sends an email to the e-commerce website by replicating a trusted user/company to gain access to their data. They mainly attempt this attack to transfer a huge amount of money to a fraudulent account.

In the proposed work, the above-discussed cyber-attacks are handled using deep learning with an optimization algorithm that avoids the vulnerability of malignant data. Deep learning optimization requires data integrity. For sophisticated deep learning optimization, the raw data requires manual procedure and examination to secure data quality with inevitable human errors are obtained. While the deep learning optimization technique processes input data, the unexpected outputs are led due to differences between raw data, and incomplete and damaged data. With the help of processed learning data required for monitoring, the various outputs of the deep learning-based HIASTLO model are verified [16].

Apply different processing and collecting services to the deep learning optimization model. Nevertheless, the major issue is information reliability and requires a traceable, analyzed, and identified data collection environment. When the middle IoT server stops or is modulated then the central server exists to manage open data. Generally, the data availability and reliability with information degradations are occurred due to serval threats including selfish mining attacks, DDoS attacks, Poisoning attacks, Sybil attacks, and Impersonation attacks. Hence, we require IoT big data server stability, and additionally, the deep learning optimization model data offer big data in the collection course. In this article, we proposed an IoT blockchain-based deep learning optimization data management method.

Thus the security analysis of the proposed method is based on five category mentioned as follows;

“The process of establishing trust in the identification of people or information systems” is known as authentication.

The phrase “critical information is not divulged to unauthorised entities” is used to describe secrecy.

Non-repudiation is the “protection against a person fraudulently denying that they have carried out a certain activity.”

“Ensuring timely and dependable access to and use of information” is the definition of availability.

Integrity is a fundamental characteristic of the blockchain technology

4.3 Deep learning optimization model

The raw data is utilized with a deep learning optimization model and must prevent the data forgery affected by a third party in terms of malicious attack. The raw data encryption must protect personal information. The proposed deep learning optimization model with IoT-based blockchain ensures and satisfies the data integrity. Before providing deep learning optimization data, the learners with data are never modulated. The IoT-based blockchain stores the hashcode of raw data, which is received from at least one IoT data provider. The raw data hashcode stored in the IoT-based blockchain compares the data hashcode used for the deep learning optimization model thereby the data integrity is verified. The accurate data provider integrity and raw data forgery preventions are enabled via raw data for the deep learning optimization model.

The raw data of block data obtained from the data provider encrypts the IoT blockchain and stores raw data. Based on the features of the deep learning optimization model, the efficiency of processing data is ensured and data integrity is provided. Further, the deep learning-based optimization model provides safety against selfish mining, DDoS, Poisoning, Sybil, and Impersonation attacks.

4.4 Deep learning model verification

The data must confirm via processing data and raw data tracking in the deep learning optimization model. The deep learning optimization model above wired and wireless networks demeanor data verification with the help of IoT-based blockchain server connection. From the data provision node, the verification node is obtained and the blockchain stores it. The raw data with its hashcode is transmitted to the data layer by the information provider layer in the E-Commerce model verification [17]. The data and hashcode obtained from the provider layer are encrypted with the usage of data collection and management. For model management, the learning patterns teach the deep learning optimization model and obtain the outputs.

Furthermore, the verification node creates the encryption and decryption keys and it saves the data obtained from the information provision node in the IoT-based blockchain. In an open network, a similar encryption key is used for the amount of data users. The hashcode obtained from the deep learning optimization model server depends upon the request for the data verification. Transmit and decrypt the encrypted hash value to the verification module.

5 Simulation results and discussion

The simulation of our method is carried out by blockchain running Hyperledger fabric v1.2 with a docker container of version 2.0.5. Virtual machine Ubuntu Linux v16.04 LTS with 1 virtual CPU core is used in our simulation. The effective execution of Hyperledge fabrics requires internal memory of about 2 GB and 20 GB of external memory. Additionally, Node.js v8.9.1 is appended to conduct the multi-node testing with the npm version of 6.8.0. Moreover, the power supply is given by using a Gigabit Ethernet switch. The experiments are carried out to analyze the user integrity authorization and cost function.

5.1 Performance analysis in terms of scalability and interoperability

The performance of scalability and interoperability are analyzed in this section. Scalability is determined as the addition of a transactional block to the chain whereas interoperability is determined as the time taken to append a block to the chain. The time of adding a block to the chain is directly proportional to the time required to create a block on the fabric network. Figure 3 illustrates the comparative study of response time for user integrity authorization with the number of transactions. We have conducted almost 1500 transactions and enlisted the data. Further, Fig. 3 shows that the response time for the user integrity authorization enhances marginally with the addition of more blocks. After the completion of 1250 transactions, the curve increases linearly corresponding to the number of transactions.

Fig. 3

Performance analysis response time for user integrity authorization vs No. of transactions.

Figure 4. shows the performance analysis of response time with respect to the number of transactions. The performance of automatic verification time and optimized data integrity time of our proposed methods were analyzed. From we can conclude that user integrity and convergence are enhanced marginally with respect to the increasing number of transactions.

Fig. 4

Performance analysis of block addition response time of auto verification and data integrity verification.

Next, the total transaction time of our proposed method is analyzed with MudraChain [19] and NormaChain [18] methods. From Fig. 5, it is evident that the proposed method transfers more data in 1 hour when compared with the other two methods.

Fig. 5

Performance analysis of total transactions time.

5.2 Performance analysis in terms of a cost function

Due to the complexity of the cost function in some optimization problems, the leading student may always be the temporary/local optimum, even though they are still far from the permanent/global optimum, and the gradual improvement of the trailing students may be constrained to a small area of design space just around them. As a result, a bad operational loop would develop, impeding the optimization process and perhaps even making it impossible to reach the global optimum. The “challenge of the new student” is intended to supplement the algorithm with ongoing rebellion against the present leader in order to give IAS a more dynamic and exploratory character. The new student will assume control of the challenge if he is more skilled than the one who is currently in the lead.

The convergence curve in Fig. 6 is used to examine the cost function of our proposed system. The global optimum was found as soon as possible, or after the 40th iteration, as can be seen in the Fig. 6. As a result, the robustness of our proposed deep learning-based HIASTLO algorithm is determined by this result. Consequently, our solution quickly makes the safe e-commerce application available.

Fig. 6

Convergence curve for the IoT based Ecommerce security applications.

5.3 Performance analysis in terms of throughput

The throughput of our proposed method is analyzed by comparing the number of IoT nodes against the total number of transactions per second. The outcome is depicted in Fig. 7 and shows that our proposed method possesses greater transactions per second and hence it fulfills the required ability for the e-commerce blockchain.

Fig. 7

Performance analysis in terms of Throughput.

5.4 Performance analysis in terms of data privacy

To analyze the performance of our proposed work in terms of data privacy we have taken some habitually applied financial datasets which are illustrated in Table 2. From Table 2 it is evident that our proposed work confirms the data privacy in the IoT-based e-commerce blockchain.

Table 2
Data privacy analysis

Frameworks Performance Data privacy analysis

Appending new customer 1. Generate a new customer application in bank A 2. Assign the disclosure 3. Make consensus protocol 4. Data stored in the smart ledger are harmonized to Bank B A new customer is appended to Bank B

1. Generate the new customer application in Bank A 2. Assign the non-closure documents 3. Make consensus protocol 4. Data stored in the smart ledger are harmonized to Bank B A new customer is cannot be appended to Bank B

Upgrading the customer 1. Select the customer possessing the disclosure file 2. Change the age details 3. Make consensus protocol 4. Data stored in the smart ledger are harmonized to Bank B New data by changing the age details are appended to the bank B

1. Select the customer who has the disclosure details in bank A but the money details are enclosed in bank C 2. Change the money and age details 3. Make consensus protocol 4. Data stored in the smart ledger are harmonized to Bank B Data stored in the smart ledger are harmonized to Bank B New data by changing the age details are appended to the bank B but bank C possess all the changed data

1. Select the customer having the non-enclosed details of Bank A 2. Change the age details 3. Make consensus protocol 4. Data stored in the smart ledger are harmonized to Bank B Data stored in the smart ledger are harmonized to Bank B The new changed details cannot be seen by bank B

1. Select the customer having the non-enclosed details of Bank A 2. Change all the enclosed details 3. Change age details 4. Make consensus protocol 5. Data stored in the smart ledger are harmonized to Bank B Data stored in the smart ledger are harmonized to Bank B Bank B can access all the details

Frameworks	Performance	Data privacy analysis
Appending new customer	1. Generate a new customer application in bank A 2. Assign the disclosure 3. Make consensus protocol 4. Data stored in the smart ledger are harmonized to Bank B	A new customer is appended to Bank B
	1. Generate the new customer application in Bank A 2. Assign the non-closure documents 3. Make consensus protocol 4. Data stored in the smart ledger are harmonized to Bank B	A new customer is cannot be appended to Bank B
Upgrading the customer	1. Select the customer possessing the disclosure file 2. Change the age details 3. Make consensus protocol 4. Data stored in the smart ledger are harmonized to Bank B	New data by changing the age details are appended to the bank B
	1. Select the customer who has the disclosure details in bank A but the money details are enclosed in bank C 2. Change the money and age details 3. Make consensus protocol 4. Data stored in the smart ledger are harmonized to Bank B Data stored in the smart ledger are harmonized to Bank B	New data by changing the age details are appended to the bank B but bank C possess all the changed data
	1. Select the customer having the non-enclosed details of Bank A 2. Change the age details 3. Make consensus protocol 4. Data stored in the smart ledger are harmonized to Bank B Data stored in the smart ledger are harmonized to Bank B	The new changed details cannot be seen by bank B
	1. Select the customer having the non-enclosed details of Bank A 2. Change all the enclosed details 3. Change age details 4. Make consensus protocol 5. Data stored in the smart ledger are harmonized to Bank B Data stored in the smart ledger are harmonized to Bank B	Bank B can access all the details

5.5 Performance Analysis in terms of cyber-attacks detection

This stage involves gathering the non-saturated data, such as speech, texts, and photos. The fitted data is then extracted for further processing after which we take care of the pre-processing. Convert the gathered data so that it can be used in the deep learning model in the event that current data properties are combined, properties are deleted or selected, and missing values are deleted or filled in. Apply the examined data to deep learning while optimising the process by removing inference and exploration data, mapping the data, and looking for standardised data patterns. The raw data collected during the data gathering procedure typically contains a few sounds that are inconsistent and are not suitable for further processing. However, increasing performance, accuracy, reliability, and quality of data is a crucial step.

Accuracy

Accuracy can be determined as the ratio of accurate predictions to the total number of IoT nodes estimated. $Accuracy = \frac{T_{P} + T_{N}}{T_{P} + T_{N} + F_{P} + F_{N}}$ (6)

Precision

Precision can be determined as the fraction of accurate predictions of data that are relevant to the process. It can be given as, $Precision P = \frac{T_{P}}{T_{P} + F_{P}}$ (7)

Recall

Recall can be determined as the successfully enabled relevant document from the fraction of accurately predicted documents. $Recall = \frac{T_{P}}{T_{P} + F_{N}}$ (8)

Here, T_P, T_N, F_N, and F_p are the true positives, true negatives, false negatives, and false positives correspondingly.

The performance metrics of our proposed deep learning-based HIASTLO is analyzed with the existing BSIoTNET [5], BCFC [6], DRNN [7], DNN-KNN [8], MOO-FS [10], LRNN [11], and HDLM [12] works. Figure 8. shows the performance analysis of our proposed method in terms of accuracy with other works mentioned above. The comparative analysis shows that our proposed method exhibits the highest accuracy of about 98.1% and the LRNN method possesses the least accuracy of about 78.21%. Thus our proposed deep learning-based HIASTLO outperforms the other approaches.

Fig. 8

Performance analysis in terms of Accuracy.

The precision is analyzed and depicted in Fig. 9. Our proposed deep learning method exhibits almost 99.16% precision, whereas, HDLM shows the least precision of about 74.25%. Moreover, our proposed method shows an almost 2.87% greater value than the existing DRNN method. The performance metrics recall is comparatively analyzed and depicted in Fig. 10. It shows that our method has 97.78% recall and the MOO-FS method shows the least recall of 70.12%. Our proposed method shows a 2.36% rise in recall than the BSIoTNET method. Therefore we can conclude that our proposed deep learning-based HIASTLO shows better performances than the other existing approaches.

Fig. 9

Performance analysis in terms of Precision.

Fig. 10

Performance analysis in terms of Recall.

5.6 Comparison in terms of several cyber attacks

Based on the several cyber attacks, the performance of our proposed deep learning-based HIASTLO algorithm is analyzed with some of the existing works such as BSIoTNET [5], BCFC [6], DRNN [7], DNN-KNN [8], MOO-FS [10], LRNN [11], and HDLM [12]. The outcomes are tabulated in Table 3. In that table, S is used to denote the algorithm protects against that particular attack and N denotes it doesn’t provide protection and if that particular attack is not considered and analyzed means it is denoted as U. Moreover, it also shows that our proposed method provides better protection against all attacks and has better security than other approaches.

Table 3
Comparison in terms of Cyberattacks

Attacks BSToTNET BCFC DRNN DNN-KNN MOO-FS LRNN HDLM Proposed

Selfish Mining S S N N S N S S

DDoS N N S S N N N S

Poisoning U N N U N S S S

Sybil N S N N S S S S

Impersonation N U S S U N N S

Attacks	BSToTNET	BCFC	DRNN	DNN-KNN	MOO-FS	LRNN	HDLM	Proposed
Selfish Mining	S	S	N	N	S	N	S	S
DDoS	N	N	S	S	N	N	N	S
Poisoning	U	N	N	U	N	S	S	S
Sybil	N	S	N	N	S	S	S	S
Impersonation	N	U	S	S	U	N	N	S

Table 4

Notation Table

S.no.	Variable	Definition
1.	T_P	True Positive
2.	T_N	True Negative
3.	F_P	False Positive
4.	F_N	False Negative
5.	L	Hidden layer output
6.	$W_{kj}^{L}$	Weight function of overall neurons j in the (L-1) ^th layer
7.	U_b and L_b	Upper and lower bound variables
8.	${HP}_{k}^{}$ and ${HP}_{j}^{}$	Initial and second trailing students
9.	ID_k and ID_j	Intrinsic competence
10.	Y _j,k new	New individual population
11.	Y _j,k old	Old individual population
12.	f_u, f_v	Fitness values

6 Conclusion

This article proposes an optimized deep learning model that can ensure data integrity in IoT-based blockchain applications. The hybrid Interactive Autodidactic School-Based Teaching-Learning Optimization (HIASTLO) algorithm is utilized for optimizing the neural networks. While ensuring data integrity, this work is also robust against cyber-attacks such as selfish mining, DDoS, poisoning, Sybil, and impersonation attacks. The throughput, scalability, and interoperability of the proposed model are effectively analyzed. The learning data integrity is verified via the IoT-based E-commerce blockchain data integrity model and thereby the reliability and confidentiality are improved. The proposed transaction time is higher compared to the Mudra chain and the Norma chain. Further, the performance of cybersecurity attack prevention is validated by using different evaluation measures such as precision, recall, and accuracy. Finally, the proposed deep learning with optimization model demonstrates without threats IoT-based blockchain.

References

Nanehkaran

Y.A.

, An introduction to electronic commerce, International Journal of Scientific & Technology Research 2(4) (2013).

Gunasekaran

, et al. E-commerce and its impact on operations management, International Journal of Production Economics 75.1 (2002), 197–185.

Zhang

and Chen

, A review of research relevant to the emerging industry trends: Industry 4.0, IoT, blockchain, and business analytics, Journal of Industrial Integration and Management 5(1) (2020), 165–180.

Puthal

, Malik

, Mohanty

S.P.

, Kougianos

and Das

, Everything You wanted to know about the blockchain: Its promise, components. Processes, and Problems, IEEE Consumer Electronics Magazine 7(4) (2018), 6–14.

Rathore

, Kwon

B.W.

and Park

J.H.

, BlockSecIoTNet: Blockchain-based decentralized security architecture for IoT network, Journal of Network and Computer Applications 143 (2019), 177–167.

Malomo , Olumide

, Danda Rawat

and Moses Garuba , Next-generation cybersecurity through a blockchain-enabled federated cloud framework, The Journal of Supercomputing 74(10) (2018), 5099–5126.

Almiani , Muder , Alia AbuGhazleh , Amer Al-Rahayfeh , Saleh Atiewi and Abdul Razaque , Deep recurrent neural network for IoT intrusion detection system, Simulation Modelling Practice and Theory 101 (2020), 102031.

De Souza , Cristiano Antonio , Carlos Becker Westphall , Renato BobsinMachado , João Bosco Mangueira Sobral and Gustavo dos SantosVieira , Hybrid approach to intrusion detection in fog-based IoTenvironments, Computer Networks 180 (2020), 107417.

Verma , Abhishek and Virender Ranga , Machine learning based intrusion detection systems for IoT applications, Wireless Personal Communications 111(4) (2020), 2287–2310.

10.

Roopak , Monika , Gui Yun Tian and Jonathon Chambers , Multi-objective-based feature selection for DDoS attack detection in IoT networks, IET Networks 9(3) (2020), 127–120.

11.

Latif , Shahid , Zhuo Zou , Zeba Idrees and Jawad Ahmad , A novel attack detection scheme for the industrial internet of things using a lightweight random neural network, IEEE Access 8 (2020), 89337–89350.

12.

Popoola , Segun

, Bamidele Adebisi , Mohammad Hammoudeh , Guan Gui and Haris Gacanin , Hybrid deep learning for botnet attack detection in the internet of things networks, IEEE Internet of Things Journal (2020).

13.

Jahangiri , Milad , Mohammad Ali Hadianfard , Mohammad Amir Najafgholipour , Mehdi Jahangiri and Mohammad Reza Gerami , Interactive autodidactic school: A new metaheuristicoptimization algorithm for solving mathematical and structuraldesign optimization problems,, Computers & Structures 235 (2020), 106268.

14.

Ummidivarapu , Vinay

, Hari

, Voruganti , Tahsin Khajah and Stéphane Pierre Alain Bordas , Isogeometric shape optimization of an acoustic horn using the teaching-learning-based optimization (TLBO) algorithm, Computer Aided Geometric Design (2020), 101881.

15.

Bau , David , Jun-Yan Zhu , Hendrik Strobelt , Agata Lapedriza , Bolei Zhou and Antonio Torralba , Understanding the role of individual units in a deep neural network, Proceedings of the National Academy of Sciences (2020).

16.

Lee

and Park

, A proposal of SH-tree based data synchronization method for secure maritime cloud, Journal of the Korea Institute of Information Security & Cryptology 26(4) (2016), 929–940.

17.

Kim

and Park

, Blockchain-based data-preserving AI learning environment model for AI cybersecurity systems in IoT service environments, Applied Sciences 10(14) (2020), 4718.

18.

Liu , Chunchi , Yinhao Xiao , Vishesh Javangula , Qin Hu , Shengling Wang and Xiuzhen Cheng , Norma Chain: A blockchain-based normalized autonomous transaction settlement system for IoT-based E-commerce, IEEE Internet of Things Journal 6(3) (2018), 4680–4693.

19.

Kabra , Naman , Pronaya Bhattacharya , Sudeep Tanwar and Sudhanshu Tyagi , MudraChain: Blockchain-based framework for automated cheque clearance in financial institutions, Future Generation Computer Systems 102 (2020), 587–574.

20.

Zhou , Jiajun and Yao Xifan , Hybrid teaching–learning-based optimization of correlation-aware service composition in cloud manufacturing, The International Journal of Advanced Manufacturing Technology 91 (2017), 10.1007/s00170-017-0008-8.

21.

Dorri , Ali , Kanhere , Salil and Jurdak , Towards an Optimized BlockChain for IoT. (2017), 10.1145/3054977.3055003.

22.

Shafay

, Ahmad

R.W.

, Salah

, et al. Blockchain for deep learning: review and open challenges, Cluster Comput (2022). https://doi.org/10.1007/s10586-022-03582-7

23.

Rathore , Shailendra , Park , Jong , Chang and Hangbae , Deep learning and blockchain-empowered security framework for intelligent 5G-Enabled IoT, IEEE Access (2021), 1–1. 10.1109/ACCESS.2021.3077069.

24.

Chinnasamy

, Deepalakshmi

, Dutta

A.K.

, You

and Joshi

G.P.

, Ciphertext-Policy attribute-based encryption for cloud storage:Toward data privacy and authentication in AI-enabled IoT system, Mathematics 10(1) (2022), 68. https://doi.org/10.3390/math10010068

25.

Chinnasamy

, Vinodhini

, Praveena

, Vinothini

and Sujitha

B.B.

, Blockchain based access control and data sharing systems for smart devices, International E- Conference on Data Analytics, Intelligent Systems and Information Security & ICDIIS 2020, Pollachi, India (2020), 11–12.

26.

Chinnasamy

, Vinothini

, Arun Kumar

, Allwyn Sundarraj

and Annlin

S.V.

, Blockchain Technology in Smart-Cities,” (eds) Blockchain Technology: Applications and Challenges. Intelligent Systems Reference Library, vol 203. Springer, Cham. https://doi.org/10.1007/978-3-030-69395-4_11.

An optimized deep learning model to ensure data integrity and security in IoT based e-commerce block chain application

Abstract

Keywords

1 Introduction

2 Related works

3.1 Hybrid interactive autodidactic school-based teaching learning optimization (HIASTLO) algorithm

3.1.1 Individual training session

3.1.2 Cooperative training session

3.1.3 Teacher phase

4.1 BlockChain

4.2 Data pre-processing and analysis

4.3 Deep learning optimization model

4.4 Deep learning model verification

5 Simulation results and discussion

5.1 Performance analysis in terms of scalability and interoperability

Table 3 Comparison in terms of Cyberattacks Attacks BSToTNET BCFC DRNN DNN-KNN MOO-FS LRNN HDLM Proposed Selfish Mining S S N N S N S S DDoS N N S S N N N S Poisoning U N N U N S S S Sybil N S N N S S S S Impersonation N U S S U N N S

References

Table 3
Comparison in terms of Cyberattacks

Attacks BSToTNET BCFC DRNN DNN-KNN MOO-FS LRNN HDLM Proposed

Selfish Mining S S N N S N S S

DDoS N N S S N N N S

Poisoning U N N U N S S S

Sybil N S N N S S S S

Impersonation N U S S U N N S