JOURNAL OF INFORMATION SCIENCE AND ENGINEERING

A PREDOMINANT INTRUSION DETECTION SYSTEM

IN IIoT USING ELCG-DSA AND LWS-BiOLSTM WITH
BLOCKCHAIN
Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi 2, Dinesh Kumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

1
Raas Infotek, Delaware, USA Email: [email protected]

2
Intel, Folsom, California, USA. Email: [email protected]

3
CGI, British Columbia, Canada. Email: [email protected]

4
Surge Technology Solutions Inc, Texas, USA.

Email: [email protected]

5
Wipro, Hyderabad, India. Email: [email protected]

6
Jouf University, Sakakah, Saudi Arabia. Email: [email protected]

Abstract: The growing connectivity of Industrial Internet of Things (IIoT) systems has
increased cyber threats, necessitating early detection of intrusions. However, existing
systems often lack focus on intermediate and continuous multifactor authorization between
IIoT and Industrial Control Systems (ICS). To overcome this, an efficient IDS for IIoT using
an Exponential Linear Congruential Generator - Digital Signature Algorithm (ELCG-DSA)
and Log Wave Sigmoid-Bidirectional Once Long Short-Term Memory (LWS-BiOLSTM)
is proposed. Initially, the industry and vehicle details are registered in the blockchain
network, and the Polychoric Entropy Correlation-Tiger Hashing Algorithm (PEC-Tiger)
generates hash codes through smart contract creation. From the generated hash codes, a
partial digital signature is created by using the ELCG-DSA technique. After login, the
registered details are processed for enhancing security using Montgomery Modulo Curve
Cryptography (MMCC). Then, the details are verified by using PEC-Tiger, and if the hash
code matches, the key generation center is notified for the creation of a fully digital signature.
After verification, the Luus–Jaakola Sequence-based Pelican Optimization Algorithm (LJS-
POA) is applied for load balancing. Next, the data security is verified in the IDS training set,
in which the features are extracted from pre-processed data. Then, the Synthetic Minority
Oversampling Technique (SMOTE) is utilized for data balancing, and LWS-BiOLSTM is
implemented to classify attacks. Furthermore, the attacked data is blocked and non-attacked
data is stored in the ICS through digital signature verification. Thus the experimental results
of the proposed framework outperform the other conventional techniques by achieving
98.78% accuracy and 98.71% security level.

Keywords: Blockchain, Intrusion Detection System (IDS), Deep Learning (DL), IIoT, Hash
Code Generation, Industry Control System, and Digital Signature Creation.

*Corresponding Author Name: Basava Ramanjaneyulu Gudivaka, Corresponding Author mail:

[email protected]
 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

1. INTRODUCTION

The IIoT states the extension as well as application of the IoT in industrial sectors (Saleh,
2024). IIoT involves the integration of sensors, software, and other technologies with
industrial equipment and processes (Singh et al., 2023) to collect, analyze, and exchange
data (Rahman et al., 2023). Additionally, IIoT improves decision-making and operational
efficiency through real-time data collection, analysis, and automation (Altunay &
Albayrak, 2023). But, IIoT introduces security challenges due to the proliferation of
connected devices (Huang et al., 2023). The challenges include vulnerabilities to cyber-
attacks, data breaches, and unauthorized access (Shahjalal et al., 2022). Thus, it poses risks
to sensitive industrial systems and critical infrastructure (Jayalaxmi et al., 2023).
Consequently, there is a pressing need for the detection of intrusions (Kumar et al., 2023)
to monitor network traffic security gaps, ensuring the integrity and security of the IIoT
environment (Ruiz-Villafranca, et al., 2023a). IDS plays a significant role in preserving
the IIoT networks as well as assets’ integrity and security (Ruiz-Villafranca, et al., 2023b).

Moreover, blockchain-based IDS leverages the decentralized and immutable nature of

blockchain to enhance security (Selvarajan et al., 2023). They offer increased transparency,
trust, and resilience against cyber threats in industrial environments (Halder & Newe,
2023). Furthermore, blockchain-based IDS offers security measures, but the incorporation
of Machine Learning (ML) and Deep Learning (DL) methodologies enhances defense
mechanisms in IIoT environments (Misra et al., 2022). ML models like Random Forest
(RF) offer better accuracy in detecting anomalies but they may require significant
computational resources (Kasongo, 2021). Subsequently, DL models like Convolutional
Neural Networks (CNNs) excel in identifying complex patterns and anomalies but it
necessitates huge amounts of labeled data for training (Ellappan et al., 2023). So, to
overcome these limitations, an efficient framework is proposed for IDS in an IIoT
environment using ELCG-DSA and LWS-BiOLSTM.

1.1 Problem Statement

Several prevailing works had some limitations, which are illustrated as follows,

• Existing works do not prioritize intermediate and continuous multifactor

authorization between IIoT and ICS to prevent attacks.
• Most of the conventional approaches exhibit improper results due to the
utilization of an unbalanced dataset.
• Many of the traditional techniques have low response during the transmission of
IIoT data to the server through sensing owing to traffic congestion.
• The limitation arises in data security, which affects the reliability and
confidentiality of sensitive information in some prevailing techniques.

The major objectives of the proposed framework are depicted further,

• To mitigate attacks, the proposed framework applied ELCG-DSA for intermediate

and continuous multifactor authorization between IIoT and ICS.
• The SMOTE algorithm is implemented to effectively address the class imbalance
issue by generating synthetic samples for minority classes, enhancing performance.
 A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

• The proposed LJS-POA technique is established to address the limitation of low

response.
• The proposed framework utilized MMCC for enhancing security and making it
well-suited for modern digital security.

The outline of this paper is illustrated as follows: Section 2 indicates the literature survey,
Section 3 depicts the proposed framework, Section 4 illustrates the results and discussion,
and Section 5 concludes the paper with future scope.

2. LITERATURE SURVEY

(Rathee et al., 2022) described a blockchain-centric IDS utilizing indirect trust intended
for IIoT systems. The study used the Viterbi algorithm for the blockchain mechanism, and
Industrial Internet of Things (IIoT) systems were employed to gauge the malicious
activities’ probability in the network. The Viterbi as well as indirect methods assessed
system transparency. Results showed improved IIoT performance, with the applied
learning technique reducing computational steps. However, a limitation was that the study
did not address the potential scalability issues and real-time implementation challenges.

(Wu et al., 2021) explained a secure and scalable IIoT system using the convergence of
blockchain as well as edge computing. The study examined cloud capabilities using the
IIoT critical infrastructure method, followed by utilizing a distributed system of blockchain
as well as edge computing paradigms to secure ledger transactions. The convergence
paradigms enabled secure and scalable critical infrastructures. Results indicated higher
effectiveness values for the IIoT blockchain. However, a limitation was the IoT devices’
resource constraints, which stopped the adoption of blockchain in IIoT.

(Dat-Thinh et al., 2022) introduced a multistage IDS using the IoT. Initially, the internet
gateways together with IoT local gateways were deployed by a collaborative IDS.
Subsequently, the IoT network classified the type of each IoT device and identified the
type of attacks targeting them. Finally, the internet gateways managed the two stages.
Results demonstrated a higher accuracy rate, but the applied method faced limitations due
to computational capacity constraints.

(Vargas et al., 2021) explained the identification of security attacks in IIoT networks using
a blockchain and ML model. Initially, IoT device networks integrated preceding solutions
to develop an integral protection system. Subsequently, IIoT identified computational
capabilities, together with threat identification, and activated secure information transfer
mechanisms. A machine learning approach was then employed to contain intruders in the
IoT network. Results showed higher accuracy in calculating malicious traffic time.
However, the performance of the applied method exhibited minimal fluctuation due to load
and scalability issues.

(Salim et al., 2022) explained a blockchain-aided secure Digital Twin (DT) methodology
for detecting botnets earlier in IIoT environments. Initially, the blockchain-enabled digital
framework detected bot creation early in a smart factory environment. Subsequently, DT
 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

was considered for a group of edge devices to collect data using a deep learning algorithm.
Smart contracts authenticated data synchronized betwixt the DT and a Packet Auditor (PA)
to detect corrupt device data transmission. Results showed that improved packet inspection
yield was attained. However, the twin framework had limitations regarding precise data
collection periods and storage capacity.

(Alosaimi & Almutairi, 2023) proffered an IDS based on BoT-IoT, focusing on quick and
accurate attack prediction in IoT networks. A novel approach combining DL and three-
level algorithms was utilized. The Bot-IoT dataset was employed for data collection, and
the BoT-IoT method was extended to augment security in other IoT applications. Results
showed higher performance of indicators. However, the high dependence on IoT and its
rapid growth increased security risks, highlighting the crucial need for network security
solutions.

(Tharewal et al., 2022) described an IDS for IIoT regarding deep reinforcement learning.
Initially, the method integrated decision-making abilities to enhance the learning process.
Subsequently, the near-end strategy optimization method was applied for IIoT intrusion
detection. The DRL IDS utilized a feature selection method in terms of light GBM. Results
indicated higher accuracy rates. However, the limitation noted here was the absence of a
distributed architecture in IoT intrusion detection systems.

3. PROPOSED METHODOLOGY FOR ELCG-DSA AND LWS-

BIOLSTM BASED IDS IN IIOT

This section demonstrates the proposed IDS in IIoT using ELCG-DSA and LWS-
BiOLSTM. The proposed model’s block diagram is illustrated in Figure 1.
 A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

Figure 1: Block diagram for the proposed framework

3.1 Registration

Initially, industry and device details are collected from IIoT devices, and then, the details
are registered in the blockchain network for secure transmission. Thus, the registered
details are represented as𝜂𝜗 𝜂𝜗.

𝜂𝜗𝑁 = {𝜂𝜗1 , 𝜂𝜗2 , 𝜂𝜗3 ,⋅⋅⋅, 𝜂𝜗𝑛 }; 𝑁 = 1 → 𝑛 (1)

Where, 𝑛 is the number of registered details. At the time of registration, a Global

Unique Identifier (GUID) is assigned to each device. Once the registration is successful,
the smart contract is created between IIoT and ICS, and the created smart contract is
represented asℜ𝛽.

3.2 IPFS Hash Code Creation

 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

Based onℜ𝛽, the Interplanetary File System (IPFS) hash codes are created using the
proposed PEC-Tiger. The tiger hashing algorithm offers quick and efficient generation of
hash codes for large amounts of data and uses Euclidean distance for large data sets, but
it’s not scale-invariant, potentially resulting in the same hash code across different texts.
So instead of Euclidean distance, the proposed framework utilized Polychoric Entropy
Correlation, which eliminates the risk of the same hash code generation for different texts.
The process of the PEC-Tiger approach is illustrated as

• Firstly, the data from𝜂𝜗 are initialized with predefined constants(𝑎, 𝑏, 𝑐) and then,
ℜ𝛽 are divided into 𝑏 number of blocks, which are represented as,

ℜ𝛽𝐵 = (ℜ𝛽1 , ℜ𝛽2 , ℜ𝛽3 ,⋅⋅⋅, ℜ𝛽𝑏 ); 𝐵 = 1𝑡𝑜𝑏 (2)

• Subsequently, the Polychoric entropy correlation coefficient (𝑓 ℘ 𝑃𝐸𝐶 ) for 𝑖𝑡ℎ and
𝑗𝑡ℎ block is evaluated to ensure accurate hashing, which is explained as,

𝑓 ℘ 𝑃𝐸𝐶 = ∑𝑖,𝑗 𝛦(ℜ𝛽𝑖 ) ⋅ 𝛦(ℜ𝛽𝑗 ) ⋅ 𝑓𝑃𝐶𝐶 (ℜ𝛽𝑖 , ℜ𝛽𝑗 ) (3)

𝑠𝑖𝑛ℎ −1 (𝜌𝑝𝑒𝑟 )
𝑓𝑃𝐶𝐶 = (4)
𝜋

Where, 𝛦(ℜ𝛽)is the entropy of eachℜ𝛽𝐵 , 𝑓𝑃𝐶𝐶 is the polychoric correlation

coefficient, and 𝜌𝑝𝑒𝑟 is the Pearson correlation coefficient.

• Now, the compression function(𝑓𝐶𝑂𝑀 ) is applied to eachℜ𝛽𝐵 and the hashed

data(𝜓ℎ𝑎𝑠ℎ )is illustrated as,

𝜓ℎ𝑎𝑠ℎ (ℜ𝛽𝐵 ) = 𝑓𝐶𝑂𝑀 (ℜ𝛽𝐵 ) (5)

• Then, 𝑎, 𝑏, 𝑐are updated based on𝑓𝐶𝑂𝑀 (ℜ𝛽𝐵 ), which is depicted as,

(𝑎𝑢𝑝 , 𝑏𝑢𝑝 , 𝑐𝑢𝑝 ) = (𝑎 ⊕ 𝑓𝐶𝑂𝑀 (ℜ𝛽𝐵 ), 𝑏 ⊕ 𝑓𝐶𝑂𝑀 (ℜ𝛽𝐵 ), 𝑐 ⊕ 𝑓𝐶𝑂𝑀 (ℜ𝛽𝐵 )) (6)

Where, 𝑎𝑢𝑝 , 𝑏𝑢𝑝 , 𝑐𝑢𝑝 are the updated constants.

• Finally, 𝑎𝑢𝑝 , 𝑏𝑢𝑝 , 𝑐𝑢𝑝 are combined to derive a unique hash, which is demonstrated
as,

𝜓ℎ𝑎𝑠ℎ 𝐹𝑖𝑛𝑎𝑙 = {𝑎𝑢𝑝 , 𝑏𝑢𝑝 , 𝑐𝑢𝑝 } (7)

Where, 𝜓ℎ𝑎𝑠ℎ 𝐹𝑖𝑛𝑎𝑙 is the final hashed data. Based on𝜓ℎ𝑎𝑠ℎ 𝐹𝑖𝑛𝑎𝑙 , the
partial digital signature is created through login. Once login is successful, the user can
access the data for further usage.

3.3 Partial Digital Signature Creation

 A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

From𝜓ℎ𝑎𝑠ℎ 𝐹𝑖𝑛𝑎𝑙 , the partial digital signature is created by using the ELCG-DSA technique.
The DSA is a secure and efficient signature algorithm, but its limited key lengths make it
less suitable for specific applications. Hence to overcome this, the Exponential Linear
Congruential Generator function is applied, which offers extended periods and improved
randomness, reducing the correlation between generated keys. The process of ELCG-DSA
is illustrated as,

Key Generation: From the key generation center, a private key(𝜅𝑝𝑟𝑖 ) , and its
corresponding public key(𝜅𝑝𝑢𝑏 ) are created by using the exponential linear congruential
generator function, which is elaborated as,

𝜅𝑝𝑢𝑏 = (𝛽𝑎 𝜅𝑝𝑟𝑖 𝑒𝑥𝑝(𝜅𝑝𝑟𝑖 ) + 𝛽𝑏 𝜅𝑝𝑟𝑖 ) ⋅ 𝑚𝑜𝑑(𝜅𝑝𝑟𝑖 ) (8)

Where, 𝛽𝑎 and 𝛽𝑏 are the base parameters.

Signature Generation: After key generation, the signatures(𝜉𝑆𝑖𝑔 ) are generated using the
following representation.

𝜉𝑆𝑖𝑔 = (𝛽𝑏𝑎𝑠𝑒 𝑟 𝑚𝑜𝑑(𝜅𝑝𝑟𝑖 )) 𝑚𝑜𝑑(𝜅𝑝𝑢𝑏 ) (9)

Then, 𝜉𝑆𝑖𝑔 is the signatures.

Signature Verification: Finally, the recipient applies the signature verification

function(𝑓𝑆𝑖𝑔 𝑉𝑒𝑟 ) to the received message (𝑅𝑒(𝜂𝜗))and 𝜉𝑆𝑖𝑔 using 𝜅𝑝𝑢𝑏 . This can be
denoted as,

𝜉𝑆𝑖𝑔 𝑉𝑒𝑟 = 𝑓𝑆𝑖𝑔 𝑉𝑒𝑟 (𝑅𝑒(𝜂𝜗) , 𝜉𝑆𝑖𝑔 (𝜂𝜗)) (10)

Where, 𝜉𝑆𝑖𝑔 𝑉𝑒𝑟 is the verified signature.

3.4 Data Security

After signature creation, 𝜂𝜗are sensed through login to ensure data security, which is done
by using the MMCC technique. ECC enhances security and makes it well-suited for
modern digital security needs; but, the private key generation by using random values
causes problems in private key generation. Hence, to overcome this, a Montgomery
Modulo Curve is utilized by replacing the elliptic curve, which offers efficient and secure
generation of private keys. The process of MMCC is illustrated as,

• Firstly, for a cryptographic operation, the Montgomery modulo curve is utilized,

which is depicted as,

𝐶𝑦 𝐸𝑏 2 = 𝐸𝑎 3 + 𝐶𝑥 𝐸𝑎 2 + 𝐸𝑎 (11)
 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

Where, 𝐸𝑎 and 𝐸𝑏 are the variables denoting the coordinates of

Montgomery modulo curve,𝐶𝑥 and 𝐶𝑦 are constants.

• Then,𝜅𝑝𝑟𝑖 is chosen from the key generation center, while 𝜅𝑝𝑢𝑏 is derived from𝜂𝜗
using the following representation.

𝜅𝑝𝑟𝑖 = 𝜅𝑝𝑢𝑏 ⋅ 𝛿𝑆𝑃 (12)

Where, 𝛿𝑆𝑃 is the starting point of the curve.

• Now, 𝜂𝜗is encrypted using𝜅𝑝𝑢𝑏 . Then, the resulting cipher text(𝜉ℓ) is illustrated
as,

𝜉ℓ = (𝑟𝑠𝑐 ⋅ 𝛿𝑆𝑃 + 𝜅𝑝𝑢𝑏 ) (13)

Where, 𝑟𝑠𝑐 is the random scalar value. Finally, the secured data(𝜉ℓ)are
processed for load balancing through data verification in the edge server, and the
process of data verification is represented as follows.

3.5 Data verification

The data verification is carried out in the edge server by generating hash codes using PEC-
Tiger, as explained in section 3.2. The generated hash codes are then compared with the
corresponding codes stored in IPFS to verify their consistency. If they match, the edge
server notifies the key generation center to create fully digital signatures using ELCG-
DSA, which is explained in section 3.3. Then, the fully digital signatures access the data
from the edge serve 𝜒 𝜓 for load balancing.

3.6 Load Balancing

Here, load balancing of𝜒 𝜓 is carried out by using the proposed LJS-POA. POA has faster
convergence speed and simple calculations, enabling wider solution space exploration.
However, it faces issues in random position updation, which leads to local optima. So
Luus-Jaakola sequence is included to overcome this issue, which enhances the exploration
phase by systematically sampling the solution space. The process of LJS-POA is illustrated
as follows,

• Firstly, the population or the collection of individuals in 𝜒 𝜓 is randomly initialized

by using the following equation.
𝜒 𝜓 𝑖,𝑗 = ℏ𝑙𝑜𝑤𝑗 + 𝑟 ∗ (ℏ𝑢𝑝𝑝𝑗 − ℏ𝑙𝑜𝑤𝑗 ); 𝑖 = 1𝑡𝑜𝑁, 𝑗 = 1𝑡𝑜𝑀 (14)
𝑢𝑝𝑝 𝑙𝑜𝑤
Where, 𝑀indicates the problematic variables, ℏ 𝑗 and ℏ 𝑗 depicts the
upper and lower boundary of 𝑗𝑡ℎ problematic data, and 𝑟 illustrates the random
variable.
• Here, the generation of random perturbation (𝜒 𝑝𝑒𝑟 ) is done by using the Luus-
Jaakola sequence(𝛼 𝐿𝐽𝑆 ), which is illustrated as,
 A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

𝛼 𝐿𝐽𝑆 (𝜒 𝑝𝑒𝑟 ) = 𝜒 𝑝𝑒𝑟 + 𝛥 ⋅ (2 ⋅ 𝑟 − 1) (15)

𝛼 𝐿𝐽𝑆 (𝜒 𝑝𝑒𝑟 𝑃 ) = {𝜒 𝑝𝑒𝑟 1 , 𝜒 𝑝𝑒𝑟 2 , 𝜒 𝑝𝑒𝑟 3 ,⋅⋅⋅, 𝜒 𝑝𝑒𝑟 𝑝 } ; 𝑃 = 1 → 𝑝 (16)

Where,𝑝 is the number of perturbations, and 𝛥is the size of

perturbations. Next, the new position of 𝑖𝑡ℎ candidate solution(ℑ (𝜒 𝜓 𝑖,𝑗 ))is
signified as,

𝜒 𝜓 𝑖,𝑗 + 𝑟 ⋅ (𝜕𝑗 − 𝑅𝜒 𝜓 𝑖,𝑗 ) + 𝜒 𝑝𝑒𝑟 𝑖 , 𝑂𝐹𝜕 < 𝑂𝐹𝑖

𝜓
ℑ (𝜒 𝑖,𝑗 ) ={ (17)
𝜒 𝜓 𝑖,𝑗 − 𝑟 ⋅ (𝜕𝑗 − 𝑅𝜒 𝜓 𝑖,𝑗 ) + 𝜒 𝑝𝑒𝑟 𝑖 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

2 2 2
𝑂𝐹(𝜒 𝜓 ) = ∑𝑁 𝜓 𝜓 𝜓
𝑖=1 (𝜒 𝑖,1 + 𝜒 𝑖,2 +⋅⋅⋅ +𝜒 𝑖,𝑀 ) (18)

Where, 𝜕𝑗 is the position of the target, 𝑅is the parameter that can be either
1 or 2, and𝑂𝐹 is the objective function value.
Then, 𝜒 𝜓 𝑖 is updated based onℑ (𝜒 𝜓 𝑖,𝑗 ), which is outlined as,

ℑ(𝜒 𝜓 𝑖 ), ℑ(𝑂𝐹𝑖 ) < 𝑂𝐹𝑖

𝜒𝜓𝑖 = { (19)
𝜒 𝜓 𝑖 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

Where, ℑ(𝑂𝐹𝑖 )is the objective function value based on the current
position.
• Now, the new position(ℑ𝑛𝑒𝑤 (𝜒 𝜓 𝑖,𝑗 )) is evaluated based on the search for optimal
solutions; then, the behavior of each 𝜒 𝜓 𝑖,𝑗 is represented as,
•
ℑ𝑛𝑒𝑤 (𝜒 𝜓 𝑖,𝑗 ) = 𝜒 𝜓 𝑖,𝑗 + 𝐶 ⋅ (1 − 𝐼 ⁄𝐼𝑚𝑎𝑥 ()(2 ⋅ 𝑟 − 1)𝜓 𝑖,𝑗 ) (20)

Where, 𝐶is the constant value, 𝐼and 𝐼𝑚𝑎𝑥 is the current and maximum
iteration, respectively.
Finally, 𝜒 𝜓 𝑖 is updated based onℑ𝑛𝑒𝑤 (𝜒 𝜓 𝑖,𝑗 ), which is illustrated as,

ℑ𝑛𝑒𝑤 (𝜒 𝜓 𝑖 ), ℑ𝑛𝑒𝑤 (𝑂𝐹𝑖 ) < 𝑂𝐹𝑖

𝜒𝜓𝑖 = { (21)
𝜒 𝜓 𝑖 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

Then, the steps are repeated until convergence, and the balanced data is
represented by𝜎𝜁. The pseudo-code for the proposed LJS-POA is depicted as,

Pseudo-code for the proposed LJS-POA

Input: 𝜉ℓ
 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

Output: 𝜎𝜁
Begin
Initialize 𝜉ℓ𝑖,𝑗 , 𝜒 𝑝𝑒𝑟 , ℑ(𝜉ℓ𝑖,𝑗 ), ℑ𝑛𝑒𝑤 (𝜉ℓ𝑖,𝑗 ), iteration(𝐼), maximum
iteration(𝐼𝑚𝑎𝑥 ())
Set(𝐼 = 1)
While(𝐼𝑚𝑎𝑥 ())
For each 𝜉ℓdo
Initialize population
𝜉ℓ𝑖,𝑗 = ℏ𝑙𝑜𝑤𝑗 + 𝑟 ∗ (ℏ𝑢𝑝𝑝𝑗 − ℏ𝑙𝑜𝑤𝑗 )
Generate 𝜒 𝑝𝑒𝑟
𝜒 𝑝𝑒𝑟 𝑃 = {𝜒 𝑝𝑒𝑟 1 , 𝜒 𝑝𝑒𝑟 2 , 𝜒 𝑝𝑒𝑟 3 ,⋅⋅⋅, 𝜒 𝑝𝑒𝑟 𝑝 }
Calculate position
ℑ(𝜉ℓ𝑖,𝑗 ) =
𝜉ℓ𝑖,𝑗 + 𝑟 ⋅ (𝜕𝑗 − 𝑅𝜉ℓ𝑖,𝑗 ) + 𝜒 𝑝𝑒𝑟 𝑖 , 𝑂𝐹𝜕 < 𝑂𝐹𝑖
{
𝜉ℓ𝑖,𝑗 − 𝑟 ⋅ (𝜕𝑗 − 𝑅𝜉ℓ𝑖,𝑗 ) + 𝜒 𝑝𝑒𝑟 𝑖 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
Update position
ℑ(𝜉ℓ𝑖 ), ℑ(𝑂𝐹𝑖 ) < 𝑂𝐹𝑖
𝜉ℓ𝑖 = {
𝜉ℓ𝑖 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
Find new position
ℑ𝑛𝑒𝑤 (𝜉ℓ𝑖,𝑗 ) = 𝜉ℓ𝑖,𝑗 + 𝐶 ⋅ (1 − 𝐼 ⁄𝐼𝑚𝑎𝑥 ()(2 ⋅ 𝑟 −
1)𝑖,𝑗 )
Update new position
ℑ (𝜉ℓ𝑖 ), ℑ𝑛𝑒𝑤 (𝑂𝐹𝑖 ) < 𝑂𝐹𝑖
𝜉ℓ𝑖 = { 𝑛𝑒𝑤
𝜉ℓ𝑖 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
End For
End While
Return: Balanced data
End

After load balancing, 𝜎𝜁is processed for testing the data in the IDS training set, in which
the IDS training includes, big-data handling, preprocessing, data balancing, feature
extraction, feature selection, and classification.

3.7 Pre-Processing

Here, data(𝜆𝜗) from the dataset are collected, and the big data handling process is
conducted using the Hadoop Distributed File System (HDFS). HDFS efficiently stores and
manages large volumes of data by providing scalability and high throughput and the data
as depicted as𝜏𝜐. Then,𝜏𝜐is preprocessed to enhance the raw sensor data under Missing
Value Imputation (MVI), Numeralization, and Normalization for the effective detection
and classification of intrusions. This is elaborated as follows,MVI

Initially, the missing values(𝜏𝑚𝑖𝑠 ) in𝜏𝜐 are identified, which contain Not a Number (NaN),
null values, or other placeholder. Then, the nature (𝜀𝜍)of 𝜏𝑚𝑖𝑠 is determined by whether
 A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

they are Missing Completely At Random (MCAR), Missing At Random (MAR), or

Missing Not At Random (MNAR).

Next, based on𝜀𝜍, for numerical, data if𝜏𝑚𝑖𝑠 is MCAR or MAR, the mean (𝜇⃗) and median
⃗⃗⃗) imputation is considered, and if MNAR, then Regression imputation (𝑅⃗⃗) is
(𝑚
considered. Meanwhile, for categorical data, if 𝜏𝑚𝑖𝑠 is MCAR or MAR, the mode
imputation (𝑚
⃗⃗⃗𝑚𝑖 ) is considered. This can be illustrated as,
𝜏𝜐𝑁
𝜇⃗ = ∑𝑛𝑁=1 (22)
𝑛

𝑚
⃗⃗⃗ = 𝑚𝑒𝑑𝑖𝑎𝑛(𝜏𝜐) (23)

𝑅⃗⃗ = 𝑟𝑒0 + 𝑟𝑒1 𝜏𝜐𝑁1 + 𝑟𝑒2 𝜏𝜐𝑁2 +⋅⋅⋅ +𝑟𝑒𝑝 𝜏𝜐𝑝𝑁 + 𝑒𝑟𝑁 (24)

⃗⃗⃗𝑚𝑖 = 𝑚𝑜𝑑 𝑒 (𝜏𝜐)

𝑚 (25)

Where, 𝑟𝑒is the regression coefficients, 𝑝 is the number of 𝑟𝑒 , and 𝑒𝑟 is the error
⃗⃗⃗, 𝑅⃗⃗, and𝑚
term.After filling 𝜏𝑚𝑖𝑠 with calculated𝜇⃗, 𝑚 ⃗⃗⃗𝑚𝑖 , the process ensures that no biases
are introduced for effective imputation of 𝜏𝑚𝑖𝑠 . Then, the data after MVI is depicted as𝜐𝜛.

Numeralization

• Primarily from𝜐𝜛, a unique numerical label(𝜍𝐿𝐵 ) of 𝑖𝑡ℎ category(𝐶𝑎𝑡)is assigned

to transform the categorical data, which is depicted as,
𝐶𝑎𝑡𝐶 = {𝐶𝑎𝑡1 , 𝐶𝑎𝑡2 , 𝐶𝑎𝑡3 ,⋅⋅⋅, 𝐶𝑎𝑡𝑐 }; 𝐶 = 1 → 𝑐 (26)

𝜍𝐿𝐵 (𝐶𝑎𝑡𝑖 ) = 𝑖 − 1 (27)

Where, 𝑐 is the number of category in 𝜐𝜛.
• Then, the binary dummy variables are created using a one-hot encoding
function(𝑓𝑜ℎ𝑒 ) for each category, which is illustrated as,
•
1; 𝑖𝑓𝜐𝜛𝑖 ∈ 𝐶𝑎𝑡𝑖
𝑓𝑜ℎ𝑒 (𝜐𝜛𝑖 , 𝐶𝑎𝑡𝑖 ) = { (28)
0; 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

• Finally, based on the order of𝜍𝐿𝐵 , binary dummy variables are assigned to𝐶𝑎𝑡.
Then, the numeralized data is represented as𝜗𝑛𝑢𝑚 .

Normalization

• Here, 𝜗𝑛𝑢𝑚 are normalized to ensure optimal data utilization for predictive
maintenance and optimization endeavors in IIoT networks. This process is
illustrated as,
• Initially, the numerical columns(𝑛𝑐) from 𝜗𝑛𝑢𝑚 are identified, and then, the scaling
functions are applied to adjust each𝑛𝑐 in𝜗𝑛𝑢𝑚 .
 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

• Then, the scaled data(𝑠𝑐𝑎(𝜗𝑛𝑢𝑚 )) is verified whether the changes cause problems.
Finally,𝑠𝑐𝑎(𝜗𝑛𝑢𝑚 )are considered as preprocessed data and ready for further
analysis. After that, 𝑠𝑐𝑎(𝜗𝑛𝑢𝑚 )are processed for data balancing.

3.8 Data balancing

Here, 𝑠𝑐𝑎(𝜗𝑛𝑢𝑚 )are balanced by using SMOTE. This technique is applied to address the
class imbalance problem by generating synthetic samples for minority classes. This
process is demonstrated as,

Primarily, the minority class set(ℓ𝜍𝑚𝑖𝑛 ()) of 𝑠𝑐𝑎(𝜗𝑛𝑢𝑚 ) is selected, followed by the
computation of the distances between each data in ℓ𝜍𝑚𝑖𝑛 to identify its k-nearest neighbors.

Then, the sampling rate(𝑠𝑟) is selected for each𝑛 ∈ 𝑚𝑐 according to the imbalanced
proportion of𝑠𝑐𝑎(𝜗𝑛𝑢𝑚 ).

ℓ𝜍̂{𝑛1 , 𝑛2 , 𝑛3 ,⋅⋅⋅, 𝑛𝑠𝑟 }𝑚𝑖𝑛 (29)

Where, ℓ𝜍̂𝑚𝑖𝑛 is the set of selected𝑠𝑟.

Next, the synthetic instances(𝑛̂) are generated for each𝑛𝜈 ∈ ℓ𝜍̂𝑚𝑖𝑛 , using the following
representation,

𝑛̂ = 𝑟 × |𝑛 − 𝑛𝜈 | + 𝑛; 𝜈 = 1𝑡𝑜𝑠𝑟 (30)

Lastly, the balanced class labels(𝜉ℏ) are formed by combining 𝑛and𝑛̂.

3.9 Feature Selection

From𝜉ℏ, the features like destination port, forward packet length maximum, and so on are
extracted for intrusion detection as well as classification. Then, the LJS-POA is applied
for feature selection from extracted features and the process is demonstrated in section 3.6.
Moreover, the final optimal solutions are the selected features(ℓ℘) such as, Destination
port, flow duration, flow byte, and so on.

3.10 Classification

Here, ℓ℘are processed for the detection as well as classification of intrusions by using the
LWS-BiOLSTM. Bi-LSTM processes input sequences in both forward and backward
directions, reducing overfitting; however, it requires more memory, and the performance
can be sensitive to random weight initializations. To reduce memory usage, Log Wave
Sigmoid activation function and once initializer is used. Then, the classifier diagram is
depicted in Figure 2.
A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

Figure 2: Classifier diagram for the proposed LWS-BiOLSTM

• Input Layer: Initially, eachℓ℘ is represented as a vector at the time step 𝑡. Then,
the input sequence is expressed as,

𝓵℘𝑵 (𝒕) = (𝓵℘𝟏 (𝒕), 𝓵℘𝟐 (𝒕), 𝓵℘𝟑 (𝒕),⋅⋅⋅, 𝓵℘𝒏 (𝒕)) (31)

• Embedding Layer: Next, ℓ℘𝑁 (𝑡)are transformed into dense vector representation
(℘𝑑𝑣
𝑖 ) by capturing the semantic meaning of ℓ℘𝑖 , which is illustrated as,

ℓ𝐸𝑇 (𝑡) = ℓ℘𝑖 (𝑡) → ℘𝑑𝑣

𝑖 (𝑡) (32)

Where, ℓ𝐸𝑇 indicates the embedding transformation function.

• Bi-LSTM Layer: This layer consists of two LSTM layers, one processing in the
forward direction(𝛨𝐹𝑜𝑟 ) and the other in the backward direction(𝛨𝐵𝑎𝑐 ). This can
be depicted as,

𝛨𝐹𝑜𝑟 (𝑡) = 𝐿𝑆𝑇𝑀(℘𝑑𝑣 (𝑡), 𝛨𝐹𝑜𝑟 (𝑡 − 1)) (33)

𝛨𝐵𝑎𝑐 (𝑡) = 𝐿𝑆𝑇𝑀(℘𝑑𝑣 (𝑡), 𝛨𝐵𝑎𝑐 (𝑡 − 1)) (34)

Where, 𝑡 − 1represents the previous time steps.

 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

• Output Layer: Next, 𝛨𝐹𝑜𝑟 and 𝛨𝐵𝑎𝑐 are combined, which represents the output
from the Bi-LSTM layer(𝛰𝐵𝑖 ).

𝛰𝐵𝑖 = [𝛨𝐹𝑜𝑟 (𝑡); 𝛨𝐵𝑎𝑐 (𝑡)] (35)

Finally, 𝛰𝐵𝑖 passed through the fully connected layer followed by the log wave sigmoid
activation function (𝛷𝐿𝑊𝑆 ), which is represented as,

ƛℏ = 𝛷𝐿𝑊𝑆 (𝜎 ⋅ 𝛰𝐵𝑖 + 𝜛) (36)

1 2
𝛷𝐿𝑊𝑆 (𝛰𝐵𝑖 ) = (1 + 𝛰𝐵𝑖 2 )𝑒 −𝛰𝐵𝑖 (37)
1+𝑒 −𝛰𝐵𝑖

Where, ƛℏillustrates the final output from the output layer,𝜎and𝜛is the
weight and bias values, respectively. Here, 𝜛 are initialized as 1 and then,
ƛℏclassifies the IIoT data as, whether it is attacked or not and the various types
of attacks such as, Brute, Started Query Language (SQL) Injection, Cross-site
Scripting (XSS), Distributed Denial-of-Service (DDoS), PortScan, Bot,
Infiltration, File Transfer Protocol (FTP), and Secure Shell (SSH), which are
illustrated as,

ƛℏ =
{𝐵𝑟𝑢𝑡𝑒, 𝑆𝑄𝐿𝐼𝑛𝑗𝑒𝑐𝑡𝑖𝑜𝑛, 𝑋𝑆𝑆, 𝐷𝐷𝑜𝑆, 𝑃𝑜𝑟𝑡𝑆𝑐𝑎𝑛, 𝐵𝑜𝑡, 𝐼𝑛𝑓𝑖𝑡𝑟𝑎𝑡𝑖𝑜𝑛, 𝐹𝑇𝑃, 𝑆𝑆𝐻}
(38)

Attacked information is blocked, while non-attacked data is stockpiled in the

cloud. Then, the cloud validates the integrity of the data using a digital signature.
Once verified, the authenticated information is then securely transmitted to the
ICS for storage. This ensures that only authenticated and unaltered data is stored
within the ICS, enhancing data integrity and security in the IIoT environment.
The pseudo-code for the proposed LWS-BiOLSTM is illustrated as,

Pseudo-code for the proposed LWS-BiOLSTM

Input: ℓ℘
Output: ƛℏ
Begin
Initialize ℓ℘𝑁 (𝑡), ℓ𝐸𝑇 (𝑡), 𝛰𝐵𝑖 , iteration(𝐼), and maximum iteration(𝐼𝑚𝑎𝑥 ())
Set(𝐼 = 1)
While(𝐼𝑚𝑎𝑥 ())
For each ℓ℘do
Compute input sequences
𝓵℘𝑵 (𝒕) = (𝓵℘𝟏 (𝒕), 𝓵℘𝟐 (𝒕), 𝓵℘𝟑 (𝒕),⋅⋅⋅, 𝓵℘𝒏 (𝒕))
Perform embedding function
ℓ𝐸𝑇 (𝑡) = ℓ℘𝑖 (𝑡) → ℘𝑑𝑣 𝑖 (𝑡)
Estimate forward LSTM
𝛨𝐹𝑜𝑟 (𝑡) = 𝐿𝑆𝑇𝑀(℘𝑑𝑣 (𝑡), 𝛨𝐹𝑜𝑟 (𝑡 − 1))
Evaluate backward LSTM
 A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

𝛨𝐵𝑎𝑐 (𝑡) = 𝐿𝑆𝑇𝑀(℘𝑑𝑣 (𝑡), 𝛨𝐵𝑎𝑐 (𝑡 − 1))

Combine both LSTM output
𝛰𝐵𝑖 = [𝛨𝐹𝑜𝑟 (𝑡); 𝛨𝐵𝑎𝑐 (𝑡)]
Apply log wave sigmoid activation function
1
𝛷𝐿𝑊𝑆 (𝛰𝐵𝑖 ) = − 𝑙𝑜𝑔(1 + 𝑒 −𝛰𝐵𝑖 )
1+𝑒 −𝛰𝐵𝑖
Examine output layer
ƛℏ = 𝛷𝐿𝑊𝑆 (𝜎 ⋅ 𝛰𝐵𝑖 + 𝜛)
End For
End While
Return:ƛℏ =
{𝐵𝑟𝑢𝑡𝑒, 𝑆𝑄𝐿𝐼𝑛𝑗𝑒𝑐𝑡𝑖𝑜𝑛, 𝑋𝑆𝑆, 𝐷𝐷𝑜𝑆, 𝑃𝑜𝑟𝑡𝑆𝑐𝑎𝑛, 𝐵𝑜𝑡, 𝐼𝑛𝑓𝑖𝑡𝑟𝑎𝑡𝑖𝑜𝑛, 𝐹𝑇𝑃, 𝑆𝑆𝐻}
End

4. RESULT AND DISCUSSION

Here, the proposed research’s performance as well as comparative analysis is conducted

in order to prove the model’s reliability and trustworthiness. Here, the proposed model is
executed on the working platform of PYTHON.

4.1 Dataset Description

The proposed framework was assessed by utilizing the NSL-KDD dataset. In this work,
the NSL-KDD is used to perform attack detection. The NSL-KDD contains 125973
records, which hold important information about network security, information security,
and cyber-attacks. Likewise, from the whole data, 80% of the data is utilized to perform
training: whereas, the remaining 20% is allocated for testing purposes.

4.2 Performance Analysis

Here, the performance analysis of the proposed model and prevailing techniques is
conducted to demonstrate the model’s effectiveness.
 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

Figure 3: Performance validation regarding precision, recall, accuracy, and specificity

Figure 3 demonstrates the performance validation of the proposed model and existing
techniques regarding precision, recall, accuracy, and specificity. Here, the proposed LWS-
BiOLSTM obtained a high precision, recall, accuracy, and specificity of 98.78%, 98.52%,
98.78%, and 98.63%, respectively. However, the existing techniques such as Bi-LSTM,
LSTM, RNN, and DN obtained a low average precision, recall, accuracy, and specificity
of 93.51%, 93.74%, 93.36%, and 93.05%, accordingly. The proposed model employed the
Log Wave Sigmoid activation function to reduce memory usage; thus, the effectiveness of
the model was shown.

Table 1: Comparison by FPR and FNR

Techniques FPR FNR

Proposed LWS-BiOLSTM 1.5124 1.5421

Bi-LSTM 3.6632 3.7125

LSTM 5.1058 4.9452

RNN 8.4107 6.6307

DNN 10.6307 9.5245

Table 1 displays the comparison by False Positive Rate (FPR) and False Negative Rate
(FNR) of the proposed LWS-BiOLSTM and prevailing techniques. Here, the proposed
model obtained a low FPR and FNR of 1.5124 and 1.5421, correspondingly. However, the
existing models such as Bi-LSTM, LSTM, RNN, and DNN obtained a high FPR of 3.6632,
 A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

5.1058, 8.4107, and 10.6307, respectively. Likewise, the existing models obtained high
FNR values. Thus, the outcomes displayed that the proposed model is better at detecting
intrusions due to the usage of the Log Wave Sigmoid activation function and once
initializer.

Figure 4: Hashcode generation time analysis

Hashcode generation time analysis of the proposed model and conventional methods such
as Tiger, whirlpool, Secure Hashing Algorithm 512 (SHA512), and Message Digest
Algorithm 5 (MD5) are shown in Figure 4. Here, the proposed PEC-Tiger algorithm took
a less hashcode generation time of 1254 ms. However, the conventional techniques took a
high average hashcode generation time of 2610 ms. Thus, the proposed model used
Polychoric Entropy Correlation for efficient hashcode generation.
 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

Figure 5: Graphical representation of the proposed model and existing techniques

Figure 5 shows the graphical demonstration of the proposed system and existing
techniques such as DSA, Simple DSA (SDSA), Qualified DSA (QDSA), and Elliptic
Curve DSA (ECDSA) with respect to digital signature creation time as well as digital
signature verification time. Here, the proposed methodology obtained better outcomes than
the existing techniques. The proposed ELCG-DSA achieved a low digital signature
creation time and digital signature verification time of 2145 ms, and 2214 ms, respectively.
However, the existing SDSA and QDSA attained a high digital signature creation time of
3365 ms and 3845 ms, accordingly. Likewise, the remaining methods attained high digital
signature creation time and digital signature verification time. Here, the proposed model
used the Exponential Linear Congruential Generator function, which effectively reduces
the correlation between generated keys; thus, the model’s effectiveness is proved.

Table 2: Performance assessment of the proposed model

Encryption Decryption Security

Techniques
Time (ms) Time (ms) Level (%)

Proposed
1267 1325 98.71
MMCC

ECC 1862 1748 96.36

RSA 2347 2236 94.19

 A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

ElGamal 2796 2841 92.14

DES 3154 3364 90.26

Performance assessment of the proposed framework and conventional techniques is shown

in Table 2. Here, the proposed MMCC obtained a low encryption and decryption time of
1267 ms and 1325 ms, accordingly. Also, the proposed MMCC achieved a security level
of 98.71%, which is higher than the conventional techniques. However, the conventional
techniques attained a high mean encryption and decryption time of 2539.75 ms and
2547.25 ms, accordingly. Also, the conventional techniques attained a low security level.
Here, the Montgomery Modulo Curve is modified with ECC to provide a secure generation
of private keys.

(a)
 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

(b)

Figure 6: Graphical representation regarding (a) response time and (b) latency

Figures 6 (a) and (b) display the graphical evaluation of the proposed approach and the
prevailing techniques regarding response time and latency. Here, the proposed model
obtained a low response time and latency of 7423 ms and 5698 for 300 numbers of data.
However, the prevailing Whale Optimization Algorithm (WOA) attained a high response
time and latency of 11247 ms and 9481 for 300 numbers of data. Likewise, other prevailing
methods obtained a high response time and latency. Here, the proposed model excellently
performs load balancing with the help of the Luus-Jaakola sequence with POA.

Table 3: Comparative assessment regarding load balancing

Technique / Number of data 100 200 300 400 500

Proposed LJS-POA 1875 2786 3865 4794 5896

POA 3845 4785 5698 6895 7845

WOA 5884 6847 7481 8748 9654

DOA 7784 8475 9856 10658 11547

OOA 9912 10254 11574 12547 13659

 A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

Table 3 displays the comparative assessment of the proposed model and prevailing
methods such as POA, WOA, Dingo Optimization Algorithm (DOA), and Osprey
Optimization Algorithm (OOA) regarding load balancing. Here, the proposed model had
a load balancing of 5896 for 500 numbers of data. But, the prevailing methods had a load
balancing of 7845, 9654, 11547, and 13659, accordingly, for 500 numbers of data. Thus,
the proposed model is superior to the prevailing methodologies due to the usage of the
Luus-Jaakola sequence.

4.3 Comparative Analysis

In this section, the comparative evaluation of the proposed model and related works are
done to prove the model’s reliability.

Table 4: Comparative analysis

Author’s name Methods Accuracy (%) Recall (%)

Proposed model LWS-BiOLSTM 98.78 98.52

(Li et al., 2020) Multi-CNN 81.33 56.07

(Awotunde et al., 2023) Ensemble classifier 93.4 92.8

(Long et al., 2023) RCLN 97.2 96.8

(Gaber et al., 2023) PSO and BA 97.83 97.7

(Rashid et al., 2023) FL 93.5 93

Table 4 shows the comparative evaluation of the proposed model and related works
regarding accuracy, and recall. Here, the proposed model obtained a high accuracy and
recall of 98.78% and 98.52%, accordingly due to the usage of the Log Wave Sigmoid
activation function and once initializer. However, the existing Multi Convolutional Neural
Network (Multi-CNN) and Ensemble classifier obtained low accuracy of 81.33% and
93.4%. Likewise, the existing Regularized Cross-layer Ladder Network (RCLN) Particle
Swarm Optimization (PSO) and Bat Algorithm (BA) obtained a low recall of 96.8% and
97.7%, respectively. Likewise, the existing Federated Learning (FL) attained low accuracy
and recall values. Thus, the outcomes proved that the proposed model is better at detecting
intrusions.

5. CONCLUSION

This paper presents an efficient IDS for IIoT utilizing ELCG-DSA and LWS-BiOLSTM
with blockchain. Here, the proposed research was implemented by utilizing the NSL-KDD
dataset. The proposed LWS-BiOLSTM achieved high accuracy, precision, and recall of
98.78%, 98.78%, and 98.52%, respectively in detecting intrusions, which proved the high
 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

effectiveness of the model. Likewise, the proposed PEC-Tiger took a less time of 1254 ms
for hashcode generation, which was less than the existing techniques. Also, the proposed
MMCC achieved a high security level of 98.71%, which demonstrated the high-level
security of the model. Similarly, the proposed LJS-POA has a low response time of 9654
ms for 500 numbers of data. Thus, the experimental outcomes proved that the proposed
methodology was superior in detecting the intrusion.

DECLARATION:

Funding Statement:
Authors did not receive any funding.
Data Availability Statement:
No datasets were generated or analyzed during the current study
Conflict of Interest
There is no conflict of interests between the authors.
Declaration of Interests:
The authors declare that they have no known competing financial interests or personal
relationships that could have appeared to influence the work reported in this paper.
Ethics approval:
Not applicable.
Permission to reproduce material from other sources:
Yes, you can reproduce.
Clinical trial registration:
We have not harmed any human person with our research data collection, which was
gathered from an already published article
Authors’ Contributions
All authors have made equal contributions to this article.
Author Disclosure Statement
The authors declare that they have no competing interests

Future scope: In future, the efforts of this article will be extended by focusing on advanced
medical imaging modalities like Magnetic Resonance Imaging (MRI) to improve the
model’s outcomes.

REFERENCE

[1]. S. Alosaimi, S. M. Almutairi, ‘An Intrusion Detection System Using BoT-IoT’,

Applied Sciences (Switzerland), 13(9), 2023, pp. 1–15.
https://doi.org/10.3390/app13095427
[2]. H. C. Altunay, Z. Albayrak, ‘A hybrid CNN + LSTM-based intrusion detection
system for industrial IoT networks’, Engineering Science and Technology, an
International Journal, 38, 2023, pp. 1–13.
https://doi.org/10.1016/j.jestch.2022.101322
[3]. J. B. Awotunde, S. O. Folorunso, A. L. Imoize, J. O. Odunuga, C. C. Lee, C. T. Li,
D. T. Do, ‘An Ensemble Tree-Based Model for Intrusion Detection in Industrial
 A Predominant Intrusion Detection System In Iiot Using Elcg-Dsa And Lws-Biolstm With Blockchain

Internet of Things Networks’, Applied Sciences (Switzerland), 13(4), 2023, pp. 1–

22. https://doi.org/10.3390/app13042479
[4]. N. Dat-Thinh, H. Xuan-Ninh, L. Kim-Hung, ‘MidSiot: A Multistage Intrusion
Detection System for Internet of Things’, Wireless Communications and Mobile
Computing, 2022, pp. 1–15. https://doi.org/10.1155/2022/9173291
[5]. V. Ellappan, A. Mahendran, M. Subramanian, J. Jotheeswaran, A. O. Khadidos, A.
O. Khadidos, S. Selvarajan, ‘Sliding principal component and dynamic reward
reinforcement learning based IIoT attack detection’, Scientific Reports, 13(1),
2023, pp. 1–17. https://doi.org/10.1038/s41598-023-46746-0
[6]. T. Gaber, J. B. Awotunde, S. O. Folorunso, S. A. Ajagbe, E. Eldesouky, ‘Industrial
Internet of Things Intrusion Detection Method Using Machine Learning and
Optimization Techniques’, Wireless Communications and Mobile Computing,
2023, pp. 1–15. https://doi.org/10.1155/2023/3939895
[7]. S. Halder, T. Newe, ‘Radio fingerprinting for anomaly detection using federated
learning in LoRa-enabled Industrial Internet of Things’, Future Generation
Computer Systems, 143, 2023, pp. 322–336.
https://doi.org/10.1016/j.future.2023.01.021
[8]. H. Huang, P. Ye, M. Hu, J. Wu, ‘A multi-point collaborative DDoS defense
mechanism for IIoT environment’, Digital Communications and Networks, 9(2),
2023, pp. 590–601. https://doi.org/10.1016/j.dcan.2022.04.008
[9]. P. L. S. Jayalaxmi, R. Saha, G. Kumar, M. Alazab, M. Conti, X. Cheng, ‘PIGNUS:
A Deep Learning model for IDS in industrial internet-of-things’, Computers and
Security, 132, 2023, pp. 1–14. https://doi.org/10.1016/j.cose.2023.103315
[10]. S. M. Kasongo, ‘An advanced intrusion detection system for IIoT Based on GA and
tree based algorithms’, IEEE Access, 9, 2021, pp. 113199–113212.
https://doi.org/10.1109/ACCESS.2021.3104113
[11]. P. Kumar, R. Kumar, A. Kumar, A. A. Franklin, S. Garg, S. Singh, ‘Blockchain and
Deep Learning for Secure Communication in Digital Twin Empowered Industrial
IoT Network’, IEEE Transactions on Network Science and Engineering, 10(5),
2023, pp. 2802–2813. https://doi.org/10.1109/TNSE.2022.3191601
[12]. Y. Li, Y. Xu, Z. Liu, H. Hou, Y. Zheng, Y. Xin, Y. Zhao, L. Cui, ‘Robust detection
for network intrusion of industrial IoT based on multi-CNN fusion’, Measurement:
Journal of the International Measurement Confederation, 154, 2020, pp. 1–27.
https://doi.org/10.1016/j.measurement.2019.107450
[13]. J. Long, W. Liang, K. C. Li, Y. Wei, M. D. Marino, ‘A Regularized Cross-Layer
Ladder Network for Intrusion Detection in Industrial Internet of Things’, IEEE
Transactions on Industrial Informatics, 19(2), 2023, pp. 1747–1755.
https://doi.org/10.1109/TII.2022.3204034
[14]. S. Misra, C. Roy, T. Sauter, A. Mukherjee, J. Maiti, ‘Industrial Internet of Things
for Safety Management Applications: A Survey’, IEEE Access, 10, 2022, pp.
83415–83439. https://doi.org/10.1109/ACCESS.2022.3194166
[15]. Rahman, M. J. Islam, S. S. Band, G. Muhammad, K. Hasan, P. Tiwari, ‘Towards a
blockchain-SDN-based secure architecture for cloud computing in smart industrial
IoT’, Digital Communications and Networks, 9(2), 2023, pp. 411–421.
https://doi.org/10.1016/j.dcan.2022.11.003
[16]. M. M. Rashid, S. U. Khan, F. Eusufzai, M. A. Redwan, S. R. Sabuj, M. Elsharief,
‘A Federated Learning-Based Approach for Improving Intrusion Detection in
 Basava Ramanjaneyulu Gudivaka 1, *, Sri Harsha Grandhi DineshKumar Reddy Basani 3, Raj Kumar
Gudivaka 4, Rajya Lakshmi Gudivaka 5, M M Kamruzzaman 6

Industrial Internet of Things Networks’, Network, 3(1), 2023, pp. 158–179.

https://doi.org/10.3390/network3010008
[17]. G. Rathee, C. A. Kerrache, M. A. Ferrag, ‘A Blockchain-Based Intrusion Detection
System Using Viterbi Algorithm and Indirect Trust for IIoT Systems’, Journal of
Sensor and Actuator Networks, 11(4), 2022, pp. 1–12.
https://doi.org/10.3390/jsan11040071
[18]. S. Ruiz-Villafranca, J. Carrillo-Mondejar, J. M. Castelo Gomez, J. Roldan-Gomez,
‘MECInOT: a multi-access edge computing and industrial internet of things
emulator for the modelling and study of cybersecurity threats’, Journal of
Supercomputing, 79(11), 2023, pp. 1–32. https://doi.org/10.1007/s11227-023-
05098-2
[19]. S. Ruiz-Villafranca, J. Roldan-Gomez, J. Carrillo-Mondejar, J. M. C. Gomez, J. M.
Villalon, ‘A MEC-IIoT intelligent threat detector based on machine learning
boosted tree algorithms’, Computer Networks, 233, 2023, pp. 1–15.
https://doi.org/10.1016/j.comnet.2023.109868
[20]. M. S. Saleh, ‘Blockchain for secure and decentralized artificial intelligence in
cybersecurity: A comprehensive review’, Blockchain: Research and Applications,
2024, pp. 1–32. https://doi.org/10.1016/j.bcra.2024.100193
[21]. M. M. Salim, A. K. Comivi, T. Nurbek, H. Park, J. H. Park, ‘A Blockchain-Enabled
Secure Digital Twin Framework for Early Botnet Detection in IIoT Environment’,
Sensors, 22(16), 2022, pp. 1–25. https://doi.org/10.3390/s22166133
[22]. S. Selvarajan, G. Srivastava, A. O. Khadidos, A. O. Khadidos, M. Baza, A.
Alshehri, J. C. W. Lin, ‘An artificial intelligence lightweight blockchain security
model for security and privacy in IIoT systems’, Journal of Cloud Computing,
12(1), 2023, pp. 1–17. https://doi.org/10.1186/s13677-023-00412-y
[23]. M. Shahjalal, M. M. Islam, M. M. Alam, Y. M. Jang, ‘Implementation of a Secure
LoRaWAN System for Industrial Internet of Things Integrated With IPFS and
Blockchain’, IEEE Systems Journal, 16(4), 2022, pp. 5455–5464.
https://doi.org/10.1109/JSYST.2022.3174157
[24]. Singh, R. K. Dhanaraj, M. Akkas Ali, P. Balaji, M. Alharbi, ‘Transfer Fuzzy
Learning enabled Streebog Cryptographic Substitution Permutation based zero trust
security in IIOT’, Alexandria Engineering Journal, 81, 2023, pp. 449–459.
https://doi.org/10.1016/j.aej.2023.08.084
[25]. S. Tharewal, M. W. Ashfaque, S. S. Banu, P. Uma, S. M. Hassen, M. Shabaz,
‘Intrusion Detection System for Industrial Internet of Things Based on Deep
Reinforcement Learning’, Wireless Communications and Mobile Computing,
2022, pp. 1–8. https://doi.org/10.1155/2022/9023719
[26]. H. Vargas, C. Lozano-Garzon, G. A. Montoya, Y. Donoso, ‘Detection of security
attacks in industrial IoT networks: A blockchain and machine learning approach’,
Electronics (Switzerland), 10(21), 2021, pp. 1–18.
https://doi.org/10.3390/electronics10212662
[27]. Y. Wu, H. N. Dai, H. Wang, ‘Convergence of Blockchain and Edge Computing for
Secure and Scalable IIoT Critical Infrastructures in Industry 4.0’, IEEE Internet of
Things Journal, 8(4), 2021, pp. 2300–2317.
https://doi.org/10.1109/JIOT.2020.3025916