Received 11 May 2023, accepted 24 May 2023, date of publication 29 May 2023, date of current version 8 June 2023.

Digital Object Identifier 10.1109/ACCESS.2023.3280798

Blockchain Assisted Data Edge Verification With

Consensus Algorithm for Machine Learning
Assisted IoT
THAVAVEL VAIYAPURI 1 , (Member, IEEE), K. SHANKAR 2,3 , (Senior Member, IEEE),
SURENDRAN RAJENDRAN2 , (Member, IEEE), SACHIN KUMAR 3 , (Senior Member, IEEE),
SRIJANA ACHARYA 4 , AND HYUNIL KIM 4
1 College of Computer Engineering and Sciences, Prince Sattam bin Abdulaziz University, Al-Kharj 11942, Saudi Arabia
2 Department of Computer Science and Engineering, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai 602105,
India
3 Big Data and Machine Learning Laboratory, South Ural State University, 454080 Chelyabinsk, Russia
4 Department of Convergence Science, Kongju National University, Chungcheongnam-do, Gongju-si 32588, South Korea

Corresponding author: Hyunil Kim ([email protected])

This work was supported in part by the Institute of Information and Communications Technology Planning and Evaluation (IITP) Grant
funded by the Korea Government (MSIT) (Development of User Identity Certification and Management Technology for Self-Sovereign
Identity Applications and Robust AI and Distributed Attack Detection for Edge AI Security) under Grant 2021-0-00565 and Grant
2021-0-00511.

ABSTRACT Internet of Things (IoT) devices are becoming increasingly ubiquitous in daily life. They are
utilized in various sectors like healthcare, manufacturing, and transportation. The main challenges related
to IoT devices are the potential for faults to occur and their reliability. In classical IoT fault detection, the
client device must upload raw information to the central server for the training model, which can reveal
sensitive business information. Blockchain (BC) technology and a fault detection algorithm are applied to
overcome these challenges. Generally, the fusion of BC technology and fault detection algorithms can give
a secure and more reliable IoT ecosystem. Therefore, this study develops a new Blockchain Assisted Data
Edge Verification with Consensus Algorithm for Machine Learning (BDEV-CAML) technique for IoT Fault
Detection purposes. The presented BDEV-CAML technique integrates the benefits of blockchain, IoT, and
ML models to enhance the IoT network’s trustworthiness, efficacy, and security. In BC technology, IoT
devices that possess a significant level of decentralized decision-making capability can attain a consensus
on the efficiency of intrablock transactions. For fault detection in the IoT network, the deep directional
gated recurrent unit (DBiGRU) model is used. Finally, the African vulture optimization algorithm (AVOA)
technique is utilized for the optimal hyperparameter tuning of the DBiGRU model, which helps in improving
the fault detection rate. A detailed set of experiments were carried out to highlight the enhanced performance
of the BDEV-CAML algorithm. The comprehensive experimental results stated the improved performance
of the BDEV-CAML technique over other existing models with maximum accuracy of 99.6%.

INDEX TERMS Blockchain, Internet of Things, consensus algorithm, fault detection, deep learning,
hyperparameter tuning.

I. INTRODUCTION more prominent as a new concept to satisfy the requirements

With the advent of 5G, wireless sensor networks (WSNs), and of ubiquitous, flexible, and agile availability of cyberspace
relevant technology, the Internet of Things (IoT) has become from physical systems [1]. But the present centralized IoT
framework is greatly limited by security challenges, single
The associate editor coordinating the review of this manuscript and point of failure, robustness, and data privacy [2]. Nowadays,
approving it for publication was Stefano Scanzio . blockchain (BC) is a potential solution to sort out such issues

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

55370 For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ VOLUME 11, 2023
T. Vaiyapuri et al.: BDEV-CAML Assisted IoT

because of its capability to maintain an immutable open II. RELATED WORKS

ledger that is easily accessed by all but is tamper-proof [3]. Trivedi et al. [11] devised a DL and BC-based EV fault
As well a wide range of innovative IoT applications is enabled detection (FD) structure to find different kinds of faults: bat-
by the rapid development of edge computing. Few research tery faults, air tire pressure, and vehicle temperature. Further,
works are done on this subject, and various tools, like BC to achieve the FD data transaction with high reliability and
and IoT, are utilized to make an intelligent supply chain. Still, scalability for EV, the author uses a 5G wireless network,
research gaps in this area are found, like the study of relations including an interplanetary file system (IPFS). Initially, the
of these three areas: BC, smart supply chain, and IoT [4]. author uses an LSTM and a CNN technique to deal with
Another application is the absence of categorization of IoT battery fault detection, air tire pressure fault, and anomaly
features and BC that disturbs building an intelligent supply detection for temperature fault to forecast the faulty dataset,
chain. BC is a unique e-book, a kind of data and reporting guaranteeing users a safer journey. In [12], the author target
mechanism that records its worth [5]. The only difference to detect a technique for potential FD in IoT gadgets. Based
between BC and other mechanisms is that the saved data is on BC, an IoT network architecture is initially developed,
shared between network members. Encryption usage makes it and a data edge authentication system is designed; the BC
almost impossible to manipulate or delete the recorded data. is utilized for assuring that data could not be tampered with
BC technology presents immense benefits like stability, and the precision can be guaranteed. Eventfully, a dataset
trust, speed, effectiveness, precision and independence [6]. accuracy-weighted RF-related PSO was devised.
The unique features of a BC program (unchangeable Belhadi et al. [13] developed an innovative structure to
records, peer-to-peer relationship, approval schedule, and precisely find anomalous patterns in privacy RL in a het-
autonomous) lead to an error elimination of transfer, erogeneous and distributed energy environment. The local
increased productivity, the use of data, data security assur- outlier factor was accomplished for deriving local abnormal
ance and time and money saving [7]. Decentralization indi- patterns in all sites of the distributed energy environment.
cates the absence of central authority or intermediaries; With BC technology, reinforcement privacy learning com-
the selected or individual participants in one BC can val- bines local anomalous patterns into global complicated ones.
idate the reports of their trading partners and be suitable Huang et al. [14] presented a Gaussian Bernoulli restricted
to use the whole database and entire history without the Boltzmann machine (GBRBM)-oriented DNN innovative
intermediary’s help [8]. Further, by eliminating the role of algorithm for transforming the FD into a classifier issue.
management intermediary, BC eradicates the necessities of The presented technique outpaces other baseline ML tech-
centralized management; the distributed system called Peer- niques by real trace-driven experiments. In [15], an innovative
to-peer systems composed of personal computers, which Energy-Efficient Heterogeneous Fault Management system
offer their computational resources (information distribution, was devised to achieve such heterogeneous faults in IWSN.
storage capacity) directly to others [9]. Indeed, BC technol- The three new diagnosis techniques can achieve efficient
ogy is devised in dispersed systems to alleviate any single heterogeneous FD in this technique. The Tuned SVM clas-
point of failure; customers need not believe their service sifier facilitated the classification of the heterogeneous faults
providers. In the industrial Internet of Things (IIoT), one where the tuning parameter of the presented method is opti-
significant problem is Fault detection. In traditional device mized using the Hierarchy-related Grasshopper Optimization
failure recognition of IIoT, for centralised model training, method.
client devices should upload local raw information to a cen- Mittra et al. [16] emphases on investigating these main
tral server [10]. This may cause problems like data privacy issues in the secondary transmission system, proposing tech-
since the local data of clients can be business sensitive. For niques to implement and integrate modern technologies like
instance, using air conditioning in hotels possibly reflects the BC and IoT, and sightseeing the scope of enhancements in the
occupancy rate. current system. The research workers have modernised con-
This study develops a new Blockchain Assisted Data ventional techniques and re-applied devices and equipment.
Edge Verification with Consensus Algorithm for Machine Zhang et al. [17] present a new adaptive privacy-preserving
Learning (BDEV-CAML) technique for IoT Fault Detection federated learning method called AdaPFL for FD in IoS,
purposes. In BC technology, IoT devices that possess a sig- which organizes various shipping agents to develop a tech-
nificant level of decentralized decision-making ability can nique by sharing model parameters without data leakage.
reach an agreement regarding the effectiveness of intrablock Initially, the author uses two tasks as instances to show that
transactions. The deep directional gated recurrent unit (DBi- a smaller portion of model parameters may expose the raw
GRU) model is used for fault detection in the IoT network. data of shipping agents. To protect the basic information of
Finally, the African vulture optimization algorithm (AVOA) shipping agents, the Paillier-based communication technique
technique is utilized for the optimal hyperparameter tuning was devised.
of the DBiGRU model, which helps in improving the fault In [18], a maintenance-oriented KG is presented ini-
detection rate. A detailed set of experiments were carried out tialy dependent upon a explicit domain oriented ontology
to highlight the enhanced performance of the BDEV-CAML scheme and accumulated preserved data. Afterward, an
algorithm. Attention-Based Compressed Relational GCN was presented

VOLUME 11, 2023 55371

 T. Vaiyapuri et al.: BDEV-CAML Assisted IoT

for predicting possible solutions and describe fault in preserve and stored in the block. The stored information is now avail-
tasks. Xia et al. [19] presents a novel hypergraph convolu- able within the distributed network, with a participant node to
tion network (HGCN) based model to forecast MRR from authenticate the transaction. Every node in BC is connected
the CMP procedure. A major contributions contains: (1) a and assists the transaction and crypto code. An additional
generic hypergraph method for representing the interactions feature in BC technology is the mathematical algorithm that
of difficult tools; and (2) a temporal enabled forecast method is extremely powerful in these networks. It is responsible for
for learning the difficult data correlation and high order rep- blocking authentication to minor nodes without affecting the
resentation dependent upon the hypergraph. In [20], a vertical data; for that reason, BC is transparent and secure. There
federated learning (FL) method, privacy-preserving boosting exist recommendation systems based on BC and knowledge
tree was established to collaborative fault analysis of indus- discovery technology and many research requirements for
trial practitioners but maintained anonymity. tackling security challenges. This procedure must implement
Xia et al. [21] examines a residual-HGCN (Res-HGCN) the incorporation of IoT and BC. Likewise, the security
technique which holistically drive in tools structure and oper- challenges, which the research workers state, make the BC
ational processes as a hypergraph procedure as data-driven a better choice. BC’s main characteristics are programma-
method allowing for reaction amongst equipment’s mecha- bility, trust, security, and so on. A BC comprises a private
nisms. Keung et al. [22] introduces the execution of devel- BC, a consortium BC, or a public BC. The public BC is
oping ARP for IIoT and resource synchronisation flexible well-known for digital currency. The primary goal of the con-
robotic and facility control scheme for addressing this chal- sortium BC is to integrate the service trading and stakeholder
lenge. In [23], the authors propose for addressing the value entity.
construction of exploiting the IIoT-driven resource synchro-
nization and sharing-based robotic mobile fulfillment system B. IoT NETWORK MODEL AND DISTRIBUTED
(RMFS) for enhancing the entire operational efficiency and LEDGER DEPLOYMENT
effectiveness in the data transmission and synchronization of If the edge gateway fails, IoT using single-edge gateways
resources. is prone to single-point failure since the whole IoT will be
The number of parameters of DL methods increases disconnected. Multiple-edge gateways decrease the delay and
quickly because of the continuous deepening of the method distance of communication of information and prevent energy
that give rise to model overfitting. Meanwhile, various hyper- loss based on the fast consumption of the IoT node nearby
parameters have a huge impact on the ability of the CNN the edge gateway. In the presented method, multiple edge
method. To be specific, the hyperparameters like learning rate gateways are the edge server and are applied to authenticate
selection, epoch count, and batch size are vital to get produc- the information received from the BC [12]. The edge node
tive results. As the trial and error method for hyperparame- process the received statistics and encapsulate the processed
ter tuning is a erroneous and tedious process, metaheuristic information, timestamp, and other transaction data into the
methods are adopted. Hence, the authors have used AVOA block that can be represented as the data block. Data authen-
method for the parameter selection of the DBiGRU approach tication can be accomplished by transporting these blocks.
in this study. Each edge gateway retains a synchronized and shared dis-
tributed ledge. Every transaction between the edge gateways
is stored in the ledger, like data exchange or asset records.
III. THE PROPOSED MODEL
Due to the computing power, limited storage capacity, energy,
In this study, we have designed a novel BDEV-CAML algo-
etc., the distributed ledger is positioned on edge gateways and
rithm to identify faults in the IoT environment effectively.
is retained by the edge gateway. The information transferred
The BDEV-CAML technique integrated the advantages of the
in IoT is stored in the ledger with decentralised features,
blockchain, IoT, and ML concepts for boosting the IoT net-
guaranteeing that the information is tamper-proof.
work’s trustworthiness, efficacy, and security. In the context
Data Consensus Algorithm: In BC technology, nodes
of intrablock transactions, IoT devices with strong decentral-
equipped with extensive decentralized decision-making abil-
ized decision-making abilities can agree on the effectiveness.
ities could attain a consensus on the efficiency of intrablock
Fig. 1 represents the overall procedure of the BDEV-CAML
transactions. To guarantee consensus amongst edge gateways,
approach.
the conventional BC technique depends mainly on the com-
puting power of distributed edge gateway. With compara-
A. BLOCKCHAIN TECHNOLOGY tively poor computation power, this doesn’t apply to IoT.
Blockchain (BC) technology is the eminent domain for safety Consequently, the study developed a data consensus model
and trust, which can be applied to any relevant topic to of IoT. The hash function is performed for transforming
keep data and information private [24]. Likewise, it is a the information into respective hash values once the edge
groundbreaking technology for distributed and decentralized gateway receives the information [12]. Every evaluated hash
computing frameworks that support the data with encrypted value corresponds to the data point, and the new information
blocks in the chain. Digital data related to time, date amount, could not be recovered through hashing. The edge gateway
transactions, and so on are enlarged in the transaction process receives the information store, the destination edge gateway,

55372 VOLUME 11, 2023

T. Vaiyapuri et al.: BDEV-CAML Assisted IoT

and other related data in the block. The data block is transmit- ht−1 and xt , which can be evaluated by Eq. (4). Consequently,
ted to each edge gateway for authentication and waits for the a candidate vector, h̃t , is generated using tanh function where
authentication outcomes. The information is authenticated by the output of reset gate, rt , only influence ht−1 . Moreover,
checking the ledger once the edge gateway receives the infor- the update gate controls the influence of the preceding state,
mation from other edge gateways. The edge gateway waits for ht−1 , and candidate vector, h, on state vector, ht , evaluated by
confirmation and then gives feedback on the authentication Eqs. (6) & (7).
outcome (correct or incorrect). The received information is
rt = σ Wr · ht−1′ xt + br
 

(4)
processed based on the authentication result from diverse
edge gateways accountable for confirmation. Based on BC h̃t = tanh (Wh · [rt Oht−1 , xt ] + bh ) (5)
technology, if the majority of the confirmed edge gateway ut = σ Wu · ht−1′ xt + bu
 

(6)
passes authentication (the confirmation outcome is true), the
ht = (1 − ut ) ht−1 + ut h̃t (7)
information is labelled as correct otherwise, they are labelled
as incorrect. Lastly, the processed information is transmitted where ht−1 and ht epitomize prior cell state and existing cell
to the management system to detect faults. state; h̃t represents a candidate activation vector; rt and ut
characterize the output of reset and update gates, correspond-
C. FAULT DETECTION USING DBIGRU MODEL ingly; and br , bh , and bu denote bias vectors; Wr , Wh , and
For fault detection in the IoT network, the AVOA with Wu characterize weight matrixes, ⊙ signifies the Hadamard
DBiGRU technique is utilized. Unlike the standard uni- product.
directional RNNs, Bi-directional network includes two
single hidden RNN layers (the backward and forward lay- D. HYPERPARAMETER TUNING USING AVOA
ers) in its framework [25]. Every layer is interconnected Finally, the AVOA technique is utilized for the opti-
with input and output layers, correspondingly. The bidi- mal hyperparameter tuning of the DBiGRU model, which
rectional model enables its network to learn the tourist helps improve the fault detection rate. The hyperparameters
volume sequence from the future and past directions. involved are learning rate, batch size, and number of epochs.
Backward and forward layers in the network read input The AVOA is motivated by the navigational and foraging
series x(x1 , x2 , x3 , . . . , xn−1 , xn ) from two opposite direc- behaviours of African vultures [26]. The biological nature of
tions, in which x forward = (x1 , x2 , x3 , . . . , xn−1 , xn ) and vultures with reference to competing and searching for food is
x backward = (xn , xn−1 , . . . xt . . . , x2 , x1 ), then attain a considered in four distinct phases. Consider N vultures in the
forward hidden state h⃗t (h⃗1 , h⃗2 , . . . , h⃗n−1 , h⃗n ) and back- environment representing the population n = {1, 2, . . . , N }.
← ← ← ← ←
ward hidden state h t ( h 1 , h 2 , . . . , h n−1 , h n ), correspond- Next, calculate the fitness function of every location position.
ingly (Eqs. (1) and (2)). Consequently, backwards and for- Where pn is the probability of choosing the 1st or 2nd group
ward series are concentrated and produce the final output that can be evaluated by
series y(y1 , y2 , . . . yt . . . , yn−1 , yn ) that can be evaluated by Fn
the following expression. pn = PN (8)
  n=1 Fn
h⃗t = f wx h⃗ · xt + Wh⃗h⃗ · h⃗ + bh⃗ (1) In Eq. (8), Fn denotes the fitness function of the nth loca-
←
  tion. Next, the construction of the 1st and 2nd groups in every
h t = f W ← ← ·xt + W←← · ht+1 + b← (2) iteration can be attained using Eq. (9):
xh hh h
←
yt = wyh⃗ · h⃗t + w
(
← · h t + by (3) first group, pn = L1
yh R (it) = (9)
← second group, pn = L2
where b← , and by denotes the respective bias vector, h⃗t , h t
h The satiated vulture with adequate energy can be moved
denote forward and backward propagation, correspondingly;
f represents a nonlinear activation function (viz., sigmoid to longer distances for searching food, while the hungry one
function); and wx h⃗ , Wh⃗h⃗ , W ← , W←← , W←← , and W ← char- cannot. Where the ranges of L1 and L2 are 0 ≤ L1 , L2 ≤ 1 and
xh hh hy xh L1 + L2 = 1. The rate of being hungry or satiated defines the
acterize the respective weight coefficient.
movement from the exploration stage to the exploitation stage
GRU cell was employed for adding to the abovementioned
that can be formulated by using Eq. (10) and (11):
bi-directional network that is a variant of LSTM cell and  
called an improved version of RNN cell. Though GRU was it
A = (2 × rand1 + 1) × x × 1 − +y (10)
barely used for tourist volume prediction, it accomplished the ITmax
desired forecast effects the same as LSTM in other time-series 5 5
     
it it
forecasts. GRU simplifies the gating model from the 3 LSTM y = h × sinz × + cos × −1
2 ITmax 2 ITmax
gates: forget, input, and output. A standard GRU cell com-
(11)
prises reset and update gates to decrease the computational
cost. Fig. 2 signifies the structure of GRU. The reset gate where x, h, and rand1 denote the random integer differs from
defines what data the existing step could be accessed from −1 to 1, −2 to 2, and 0 to 1, correspondingly; A represents

VOLUME 11, 2023 55373

 T. Vaiyapuri et al.: BDEV-CAML Assisted IoT

the vulture with the highest energy, z describes the prospect TABLE 1. Precision rate analysis of BDEV-CAML approach with other
systems under varying classes.
of entering the exploration phase; it and ITmax denote the cur-
rent and maximal iteration, correspondingly. The exploration
phase defines the process of finding food by the African
vultures where the parameter p1 , 0 ≤ p1 ≤ 1 determines the
selection of strategy thus,
(
(6) , p1 ≥ rand2
P (it + 1) = (12)
(8) , p1 < rand2
P (it + 1) = R (it) − D (it) × A (13)
D (it) = |q × R (it) − P (it)| (14)
Here P(it + l) indicate the vultures’ location vector in the Bestvu1ture1 (it) × P (it)
following iteration. A and R(it) are attained, correspondingly B1 = Bestvulture1 (it) − ×A
Bestvu1ture1 (it) × P(it)2
q = 2 × rand3 , where rand3 indicates a random integer
(24)
ranging from zero to one.
Bestvu1ture2 (it) × P (it)
P (it + l) = R (it) − A + rand4 × ((ub − l) × rand5 + lb ) B2 = Bestvulture2 (it) − ×A
Bestvu1ture2 (it) × P(it)2
(15) (25)
Eq. (15), ub and 1b denote the upper and lower boundaries where Best_vulture1(it) and Best_vulture2(it) represent the
of the parameter; correspondingly, Rand4 and Rand5 show optimum vulture of the 1st and 2nd groups, correspondingly,
the random integers within [0, 1]. The exploitation phase has in the existing iteration. The aggressive competition amongst
two distinct stages. The selection of any strategy relies on the vultures is given below:
the variables p2 and p3 . p2 and p3 values lie within [0, 1].
If |F| lies within [1, 0.5], then the exploitation phase enters P(it + 1) = R(it) − |d(it)| × A × Levy(d) (26)
the initial stage that defines two strategies, namely siege fight where d1 signifies the dimension of optimization problem:
and rotating flight:
u×σ
( Levy(x) = 0.01 × ,
(10) , p2 ≥ rand6 1
P (it + 1) = (16) |v| β ′
(11) , p2 < rand6
Y (1 + β) × sin πβ
 
2 1
In Eq. (16), rand6 denotes a random integer within [0, 1]. σ =( )β (27)
β−1

The solution to Eq. (16) is shown below: Y (1 + 2β) × β × 2 2
P (it + 1) = D (it) × (A + rand7 ) − d (it) (17) Fitness selection is a critical factor in the AVOA method.
D (it) = R (it) − P (it) (18) Solution encoding is used to assess the goodness (aptitude)
of the solution candidate. Then, the accuracy value is the
Next, the rotational flight of the vulture can be modelled
primary condition applied to design a fitness function.
by:
Fitness = max (P) (28)
P (it + l) = R (it) − (M1 + M2 ) (19)
TP
rand8 × P (it)
 
P= (29)
M1 = R (it) × × cos (P (it)) (20) TP + FP
2π
From the expression, TP characterizes the true positive,
rand9 × P (it)
 
M2 = R (it) × × sin (P (it)) (21) and FP indicates the false positive value.
2π
where rand8 and rand9 denote two random integers within IV. RESULTS AND DISCUSSION
[0, 1]. if |F| > 0.5, then the exploitation stage enters the next In this section, the experimental outcomes of the BDEV-
stage that describes the two dissimilar approaches of vultures, CAML technique are studied under different measures. The
namely aggressive and accumulation fight and siege to search results are inspected under various types of faults. In Table 1,
for food. Based on the following condition, the selection of and Fig. 3, a comparative precision rate (PR) analysis of the
any strategy can be done: BDEV-CAML technique is given.
( The results indicate that the RF model attains the least
(16) , p3 ≥ rand7
P (it + 1) = (22) efficiency, whereas the PSO-DAWRF and DAWRF models
(19) , p3 < rand7 attain closer results. Nevertheless, the BDEV-CAML tech-
where nique reaches better PR values. For instance, with CPUHog
B1 + B2 fault class, the BDEV-CAML technique obtains increasing
P (it + 1) = (23) PR of 99.55% while the PSO-DAWRF, DAWRF, and RF
2
55374 VOLUME 11, 2023
T. Vaiyapuri et al.: BDEV-CAML Assisted IoT

FIGURE 3. PR analysis of BDEV-CAML approach under varying classes.

FIGURE 1. The overall procedure of the BDEV-CAML system.

FIGURE 4. RR analysis of BDEV-CAML approach under varying classes.

methods achieve a reducing RR of 98.51%, 97.55%, and

FIGURE 2. The architecture of GRU. 96.32% correspondingly. Simultaneously, with IOHog fault
class, the BDEV-CAML method accomplished maximum RR
of 99.61% while the PSO-DAWRF, DAWRF, and RF meth-
models accomplish decreasing PR of 98.6%, 97.06%, and ods attained minimum RR of 98.06%, 96.86%, and 95.51%
95.87% correspondingly. Simultaneously, with IOHog fault correspondingly.
class, the BDEV-CAML system obtains increasing PR of In Table 3, and Fig. 5, a comparative accuracy rate
99.72% while the PSO-DAWRF, DAWRF, and RF meth- (AR) analysis of the BDEV-CAML method is given. The
ods achieve decreasing PR of 97.91%, 96.41%, and 95.25% results show that the RF method accomplishes the least
respectively. efficiency, whereas the PSO-DAWRF and DAWRF methods
In Table 2, and Fig. 4, a comparative recall rate (RR) attain closer results. Nonetheless, the BDEV-CAML method
analysis of the BDEV-CAML method is given. The results reaches better AR values. For the case with CPUHog fault
show that the RF method accomplishes minimum efficiency, class, the BDEV-CAML method obtains an increasing AR of
whereas the PSO-DAWRF and DAWRF methods achieve 99.62% while the PSO-DAWRF, DAWRF, and RF methods
closer outcomes. Nonetheless, the BDEV-CAML method achieve a minimum AR of 98.44%, 96.78%, and 95.07%
obtains increasing RR values. For example, with CPUHog correspondingly. Simultaneously, with IOHog fault class, the
fault class, the BDEV-CAML technique reaches a better BDEV-CAML method obtains a maximum AR of 99.53%,
RR of 99.49% while the PSO-DAWRF, DAWRF, and RF while the PSO-DAWRF, DAWRF, and RF methods achieve a

VOLUME 11, 2023 55375

 T. Vaiyapuri et al.: BDEV-CAML Assisted IoT

FIGURE 8. Loss curve of the BDEV-CAML approach.

FIGURE 5. AR analysis of BDEV-CAML approach under varying classes.

FIGURE 9. FDA analysis of BDEV-CAML approach with other existing

methods.

FIGURE 6. FR analysis of BDEV-CAML approach under varying classes.

TABLE 2. Recall rate analysis of BDEV-CAML approach with other systems

under varying classes.

FIGURE 7. Accuracy curve of the BDEV-CAML approach.

indicate that the RF model reaches the tiniest efficiency,
whereas the PSO-DAWRF and DAWRF methods attain
minimum AR of 98.03%, 96.94%, and 95.81% correspond- closer results. Nonetheless, the BDEV-CAML technique
ingly. obtains better FR values. For example, with CPUHog fault
In Table 4, and Fig. 6, a comparative F-score rate (FR) class, the BDEV-CAML technique gets maximum FR of
analysis of the BDEV-CAML method is given. The results 99.55% while the PSO-DAWRF, DAWRF, and RF models

55376 VOLUME 11, 2023

T. Vaiyapuri et al.: BDEV-CAML Assisted IoT

TABLE 3. Accuracy rate analysis of the BDEV-CAML approach with other TABLE 5. FDA analysis of BDEV-CAML approach with other existing
systems under varying classes. methods [12].

TABLE 4. F-score rate analysis of BDEV-CAML approach with other

systems under varying classes.

BDEV-CAML technique gains an increasing FDA of 87.06%

while the PSO-DAWRF, NFD, ETXTD, and DFD meth-
ods acquire minimum FDA 83.14%, 81.38%, 74.89%, and
65.83% correspondingly. These results assured the improved
performance of the BDEV-CAML technique over other exist-
ing models.

accomplish minimum FR of 98.6%, 97.06%, and 95.87% V. CONCLUSION

correspondingly. In this study, we have designed a novel BDEV-CAML algo-
Simultaneously, with IOHog fault class, the BDEV-CAML rithm for the effectual identification of faults in the IoT
method accomplishes a maximum FR of 99.72%, while the environment. The presented BDEV-CAML technique inte-
PSO-DAWRF, DAWRF, and RF methods obtain a minimum grated the advantages of the blockchain, IoT, and ML con-
FR of 97.91%, 96.41%, and 95.25% correspondingly. cepts for boosting the trustworthiness, efficacy, and security
Fig. 7 examines the accuracy of the BDEV-CAML method of the IoT network. In BC technology, IoT devices with
during the training and validation process on the test dataset. highly decentralized decision-making capability can attain a
The figure indicates that the BDEV-CAML method accom- consensus on the efficiency of intrablock transactions. For
plishes increasing accuracy values over epochs. In addition, fault detection in the IoT network, the AVOA with DBi-
the increasing validation accuracy over training accuracy GRU technique is utilized for the optimal hyperparameter
shows that the BDEV-CAML method learns efficiently on the tuning of the DBiGRU model, which helps in improving
test dataset. the fault detection rate. A detailed set of experiments were
The loss analysis of the BDEV-CAML technique at the carried out to highlight the enhanced performance of the
time of training and validation is demonstrated on the test BDEV-CAML algorithm. The comprehensive experimental
dataset in Fig. 8. The results show that the BDEV-CAML results stated the improved performance of the BDEV-CAML
technique obtains closer training and validation loss values. technique over other existing models. In future, hybrid
Note that the BDEV-CAML method learns efficiently on the DL models can boost the performance of the BDEV-
test dataset. CAML technique. In addition, the proposed model can be
Table 5 and Fig. 9 demonstrate the fault detection accuracy extended to the detection of the faults in the real time IoT
(FDA) results of the BDEV-CAML technique with recent environment.
models [12].
The results show better outcomes of the BDEV-CAML REFERENCES
technique with increasing FDA values under all values of [1] W. Zhang, Q. Lu, Q. Yu, Z. Li, Y. Liu, S. K. Lo, S. Chen, X. Xu, and
fault probability (FP). For instance, with FP of 0.05, the L. Zhu, ‘‘Blockchain-based federated learning for device failure detection
BDEV-CAML technique gains increasing FDA of 99.10% in industrial IoT,’’ IEEE Internet Things J., vol. 8, no. 7, pp. 5926–5937,
Apr. 2021.
while the PSO-DAWRF, NFD, ETXTD, and DFD mod-
[2] W. Zhang, Z. Wang, and X. Li, ‘‘Blockchain-based decentralized federated
els obtain reducing FDA of 97.88%, 96.67%, 96.26%, and transfer learning methodology for collaborative machinery fault diagno-
94.10% respectively. Simultaneously, with FP of 0.50, the sis,’’ Rel. Eng. Syst. Saf., vol. 229, Jan. 2023, Art. no. 108885.

VOLUME 11, 2023 55377

 T. Vaiyapuri et al.: BDEV-CAML Assisted IoT

[3] J. Liu, H. Liu, C. Chakraborty, K. Yu, X. Shao, and Z. Ma, ‘‘Cascade [23] K. L. Keung, C. K. M. Lee, and P. Ji, ‘‘Industrial Internet of Things-
learning embedded vision inspection of rail fastener by using a fault detec- driven storage location assignment and order picking in a resource syn-
tion IoT vehicle,’’ IEEE Internet Things J., vol. 10, no. 4, pp. 3006–3017, chronization and sharing-based robotic mobile fulfillment system,’’ Adv.
Feb. 2023. Eng. Informat., vol. 52, Apr. 2022, Art. no. 101540.
[4] T. Mezair, Y. Djenouri, A. Belhadi, G. Srivastava, and J. C.-W. Lin, [24] E. A. Adeniyi, R. O. Ogundokun, S. Misra, J. B. Awotunde, and
‘‘A sustainable deep learning framework for fault detection in 6G indus- K. M. Abiodun, ‘‘Enhanced security and privacy issues in a multi-
try 4.0 heterogeneous data environments,’’ Comput. Commun., vol. 187, tenant environment of green computing using blockchain technology,’’ in
pp. 164–171, Apr. 2022. Blockchain Applications in the Smart Era. Cham, Switzerland: Springer,
[5] M. Ul Mehmood, A. Ulasyar, A. Khattak, K. Imran, H. S. Zad, and S. Nisar, 2022, pp. 65–83.
‘‘Cloud based IoT solution for fault detection and localization in power [25] M. Lu and Q. Xie, ‘‘A novel approach for spatially controllable high-
distribution systems,’’ Energies, vol. 13, no. 11, p. 2686, May 2020. frequency forecasts of park visitation integrating attention-based deep
[6] S. U. Jan, Y. D. Lee, and I. S. Koo, ‘‘A distributed sensor-fault detection learning methods and location-based services,’’ ISPRS Int. J. Geo-
and diagnosis framework using machine learning,’’ Inf. Sci., vol. 547, Information, vol. 12, no. 3, p. 98, Feb. 2023.
pp. 777–796, Feb. 2021. [26] S. R. Nayak, R. K. Khadanga, S. Panda, P. R. Sahu, S. Padhy, and
[7] H. Zhang, R. Li, and C. Shi, ‘‘Deep learning technology of Internet of T. S. Ustun, ‘‘Participation of renewable energy sources in the frequency
Things blockchain in distribution network faults,’’ J. Intell. Syst., vol. 31, regulation issues of a five-area hybrid power system utilizing a sine cosine-
no. 1, pp. 965–978, Aug. 2022. adopted African vulture optimization algorithm,’’ Energies, vol. 16, no. 2,
p. 926, Jan. 2023.
[8] A. H. Sodhro, A. Lakhan, S. Pirbhulal, T. M. Groenli, and H. Abie, ‘‘A
lightweight security scheme for failure detection in microservices IoT-edge
networks,’’ in Sensing Technology. Cham, Switzerland: Springer 2022,
pp. 397–409.
[9] Y. Jiahao, X. Jiang, S. Wang, K. Jiang, and X. Yu, ‘‘SVM-BiLSTM: A fault
detection method for the gas station IoT system based on deep learning,’’
IEEE Access, vol. 8, pp. 203712–203723, 2020. THAVAVEL VAIYAPURI (Member, IEEE) is cur-
[10] M. Dzaferagic, N. Marchetti, and I. Macaluso, ‘‘Fault detection and clas- rently an Assistant Professor with the College
sification in industrial IoT in case of missing sensor data,’’ IEEE Internet of Computer Engineering and Sciences, Prince
Things J., vol. 9, no. 11, pp. 8892–8900, Jun. 2022. Sattam bin Abdulaziz University. With nearly
[11] M. Trivedi, R. Kakkar, R. Gupta, S. Agrawal, S. Tanwar, V.-C. Niculescu, 20 years of research and teaching experience, she
M. S. Raboaca, F. Alqahtani, A. Saad, and A. Tolba, ‘‘Blockchain and deep has published more than 50 research papers in
learning-based fault detection framework for electric vehicles,’’ Mathemat- impacted journals and international conferences.
ics, vol. 10, no. 19, p. 3626, Oct. 2022. Her research interests include data science, secu-
[12] W. Zhang, J. Wang, G. Han, S. Huang, Y. Feng, and L. Shu, ‘‘A data set rity, computer vision, and high-performance com-
accuracy weighted random forest algorithm for IoT fault detection based puting. She is a member of the IEEE Computer
on edge computing and blockchain,’’ IEEE Internet Things J., vol. 8, no. 4,
Society. She is a fellow of HEA, U.K.
pp. 2354–2363, Feb. 2021.
[13] A. Belhadi, Y. Djenouri, G. Srivastava, A. Jolfaei, and J. C.-W. Lin,
‘‘Privacy reinforcement learning for faults detection in the smart grid,’’ Ad
Hoc Netw., vol. 119, Aug. 2021, Art. no. 102541.
[14] H. Huang, S. Ding, L. Zhao, H. Huang, L. Chen, H. Gao, and S. H. Ahmed,
‘‘Real-time fault detection for IIoT facilities using GBRBM-based DNN,’’
IEEE Internet Things J., vol. 7, no. 7, pp. 5713–5722, Jul. 2020. K. SHANKAR (Senior Member, IEEE) received
[15] S. Lavanya, A. Prasanth, S. Jayachitra, and A. Shenbagarajan, ‘‘A tuned the Ph.D. degree in computer science from
classification approach for efficient heterogeneous fault diagnosis in Alagappa University, Karaikudi, India. He is cur-
IoT-enabled WSN applications,’’ Measurement, vol. 183, Oct. 2021, rently a Postdoctoral Fellow with the Big Data
Art. no. 109771. and Machine Learning Laboratory, South Ural
[16] S. Mittra, A. Aprameya, and B. K. Mohanta, ‘‘Smart grid power theft and State University, Russia, and an Adjunct Fac-
fault detection using IoT and blockchain,’’ in Proc. Int. Conf. Advance- ulty of the Department of Computer Science
ments Electr., Electron., Commun., Comput. Autom. (ICAECA), Oct. 2021, and Engineering, Saveetha School of Engineering,
pp. 1–5. Saveetha Institute of Medical and Technical Sci-
[17] Z. Zhang, C. Guan, H. Chen, X. Yang, W. Gong, and A. Yang, ‘‘Adaptive ences, Chennai, India. He has authored/coauthored
privacy-preserving federated learning for fault diagnosis in Internet of over 150 ISI journal articles (with a total impact factor of more than 350) and
Ships,’’ IEEE Internet Things J., vol. 9, no. 9, pp. 6844–6854, May 2022. more than 100 Scopus-indexed articles. Besides, he has authored/edited six
[18] L. Xia, Y. Liang, J. Leng, and P. Zheng, ‘‘Maintenance planning recom- international books published by recognized publishers, such as Springer and
mendation of complex industrial equipment based on knowledge graph CRC. His current research interests include healthcare applications, secret
and graph neural network,’’ Rel. Eng. Syst. Saf., vol. 232, Apr. 2023, image sharing schemes, digital image security, cryptography, the Internet of
Art. no. 109068. Things, and optimization algorithms.
[19] L. Xia, P. Zheng, X. Huang, and C. Liu, ‘‘A novel hypergraph convo-
lution network-based approach for predicting the material removal rate
in chemical mechanical planarization,’’ J. Intell. Manuf., vol. 33, no. 8,
pp. 2295–2306, Dec. 2022.
[20] L. Xia, P. Zheng, J. Li, W. Tang, and X. Zhang, ‘‘Privacy-preserving gradi-
ent boosting tree: Vertical federated learning for collaborative bearing fault
diagnosis,’’ IET Collaborative Intell. Manuf., vol. 4, no. 3, pp. 208–219, SURENDRAN RAJENDRAN (Member, IEEE)
Sep. 2022. received the Ph.D. degree in computer science and
[21] L. Xia, Y. Liang, P. Zheng, and X. Huang, ‘‘Residual-hypergraph convolu- engineering from Sathyabama University, in 2014.
tion network: A model-based and data-driven integrated approach for fault He is currently a Professor with the Department
diagnosis in complex equipment,’’ IEEE Trans. Instrum. Meas., vol. 72, of Computer Science and Engineering, Saveetha
pp. 1–11, 2023. School of Engineering, Saveetha Institute of Med-
[22] K. L. Keung, Y. Y. Chan, K. K. H. Ng, S. L. Mak, C. H. Li, Y. Qin, and ical and Technical Sciences, India. His research
C. W. Yu, ‘‘Edge intelligence and agnostic robotic paradigm in resource interests include cloud computing, the Internet of
synchronisation and sharing in flexible robotic and facility control sys- Things, deep learning, artificial intelligence, and
tem,’’ Adv. Eng. Informat., vol. 52, Apr. 2022, Art. no. 101530. big data.

55378 VOLUME 11, 2023

T. Vaiyapuri et al.: BDEV-CAML Assisted IoT

SACHIN KUMAR (Senior Member, IEEE) HYUNIL KIM received the B.S. degree in applied
received the Ph.D. degree in data mining/machine mathematics and the M.S. and Ph.D. degrees
learning from the Indian Institute of Technology in information security from Kongju National
Roorkee, in 2017. He is currently with the Depart- University, South Korea, in 2014, 2016, and
ment of Computer Science, South Ural State Uni- 2019, respectively. Also, he was a Postdoctoral
versity, Chelyabinsk, Russia. He is also the Head Researcher at the Daegu Gyeongbuk Institute of
of the Data Mining and Virtualization Labora- Science and Technology (DGIST), Daegu, South
tory and a leading Researcher with the Big Data Korea. He is currently a Research Professor with
and Machine Learning Research Laboratory. His Kongju National University. His research interests
research interests include intelligent transportation include AI security with federated learning, decen-
systems, machine learning, data mining, the IoT, and health informatics. tralized identifiers, and blockchain.
He also served as a reviewer for various reputed international journals.

SRIJANA ACHARYA received the B.S. degree

in computer application from MCRP University,
India, in 2003, and the M.S. degree in infor-
mation and communication engineering and the
Ph.D. degree in digital convergence business from
Yeungnam University, in 2014 and 2021, respec-
tively. She received various scholarships for her
M.S. and Ph.D. studies. She is currently a Post-
doctoral Researcher with the Department of Con-
vergence Science, Kongju National University,
South Korea. Her research interests include webometrics, open data, data
security, SNS security, SNS analysis, knowledge management, and digital
convergence.

VOLUME 11, 2023 55379