Content uploaded by Radek Martinek
Author content
All content in this area was uploaded by Radek Martinek on May 04, 2022
Content may be subject to copyright.
2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3165945, IEEE Journal of
Biomedical and Health Informatics
1
Federated-Learning Based Privacy Preservation and
Fraud-Enabled Blockchain IoMT System for
Healthcare
Abdullah lakhan ID , Mazin Abed Mohammed, Jan Nedoma, Radek Martinek, Prayag Tiwari ID , Ankit Vidyarthi ID ,
Ahmed Alkhayyat, and Weiyu Wang
Abstract—These days, the usage of machine-learning-enabled
dynamic Internet of Medical Things (IoMT) systems with mul-
tiple technologies for digital healthcare applications has been
growing progressively in practice. Machine learning plays a
vital role in the IoMT system to balance the load between
delay and energy. However, the traditional learning models fraud
on the data in the distributed IoMT system for healthcare
applications are still a critical research problem in practice.
The study devises a federated learning-based blockchain-enabled
task scheduling (FL-BETS) framework with different dynamic
heuristics. The study considers the different healthcare applica-
tions that have both hard constraint (e.g., deadline) and resource
energy consumption (e.g., soft constraint) during execution on
the distributed fog and cloud nodes. The goal of FL-BETS is to
identify and ensure the privacy preservation and fraud of data
at various levels, such as local fog nodes and remote clouds,
with minimum energy consumption and delay, and to satisfy
the deadlines of healthcare workloads. The study introduces
the mathematical model. In the performance evaluation, FL-
BETS outperforms all existing machine learning and blockchain
mechanisms in fraud analysis, data validation, energy and delay
constraints for healthcare applications.
Index Terms—IoMT, Federated Learning, Fraud-Analysis,
Blockchain, Fog, Cloud, Healthcare, Federated Learning, Privacy
Preservation.
I. INT ROD UC TI ON
The utilization of digital healthcare applications based on
the Internet of Medical Things (IoMT) system has been
increasing day by day [1]. The IoMT system collects different
Funding: This research work was partially supported by Ministry of
Education of Czech Republic (project No. SP2022/18 and SP2022/34).
(Corresponding author: Prayag Tiwari, Ankit Vidyarthi)
A. Lakhan is with the College of Computer Science and Artificial In-
telligence, Wenzhou University, Wenzhou 325035, China; E-mail: abdul-
lah@seu.edu.cn
M.A. Mohammed is with the College of Computer Science and Information
Technology, University of Anbar, Anbar 31001, Iraq; Email: mazinalshu-
jeary@uoanbar.edu.iq
J. Nedoma is with the Department of Telecommunications, VSB-
Technical University of Ostrava, 70800 Ostrava, Czech Republic; Email:
jan.nedoma@vsb.cz
R. Martinek is with the Department of Cybernetics and Biomedical
Engineering, VSB-Technical University of Ostrava , 70800 Ostrava, Czech
Republic; Email: radek.martinek@vsb.cz
P. Tiwari is with the Department of Computer Science, Aalto University,
Finland; Email: prayag.tiwari@ieee.org
A. Vidyarthi is with the Department of CSE&IT, Jaypee Institute of
Information Technology Noida, India; Email: dr.ankit.vidyarthi@gmail.com
A. Alkhayyat is with the College of technical engineering, The Islamic
University, Najaf, Iraq; Email: ahmedalkhayyat85@iunajaf.edu.iq
W. Wang is with the Business School of Changzhou University, China;
Email: weiyuwang001@gmail.com
medical devices, wireless technologies, and fog and cloud
nodes distributed throughout the network. The IoMT offers
different services to digital healthcare applications and is gen-
erally called IoT-enabled digital healthcare applications. This
digital healthcare consists of other applications in which IoT-
enabled services provide ubiquitous connectivity to the users
to monitor their healthcare 24/7. The digital healthcare IoT
applications store and migrate user data via different connected
nodes such as wireless technologies and fog and cloud comput-
ing nodes for processing. Artificial intelligence-enabled many
dynamic methods help applications manage their execution
and data storage in the IoT fog cloud network. Furthermore,
existing artificial intelligence techniques are proposed to deal
with the privacy preservation and fraud anomaly detection
issues in the network. IoT applications with blockchain pursue
various anomaly detection techniques on transactional network
data of a public financial blockchain called bitcoin [2]. This
blockchain-enabled solution is a prototype for a study that
examines anomaly detection in the context of blockchain
technology and its financial applications. It uses unsupervised
machine learning techniques to remove transactional data from
the bitcoin blockchain and analyze it for malicious transactions
[3].
Federated learning is an artificial intelligence paradigm that
concedes computing nodes to find out from a shared model
collaboratively. It works by allowing individual model training
on separate, independent IoMT data of applications while only
sharing the trained models, which do not contain any personal
data. The user devices train their applications data locally
and shared to the global computing model for execution.
This process is repeated for several iterations until a high-
quality model is generated [3]. The decentralized blockchain
is the technology that encourages the IoT applications to
execute at different nodes in the fog-cloud network [4]. The
blockchain technology can be implemented at the client-side
and server-side with different schemes such as smart-contract,
miners, consensus, and different fault-tolerant schemes [5].
However, static rules-based blockchain still suffers from dy-
namic frauds and scams, and the static learning process in
existing blockchain systems did not work efficiently in the
dynamic environment.
This paper formulates federated learning-based privacy pre-
serving and fraud detection-enable blockchain IoMT system
for healthcare applications in fog-cloud assisted network. The
study suggests a new system that leverages different technolo-
Authorized licensed use limited to: Southeast University. Downloaded on April 09,2022 at 00:17:01 UTC from IEEE Xplore. Restrictions apply.
2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3165945, IEEE Journal of
Biomedical and Health Informatics
2
gies such as federated learning and blockchain technology
is a decentralized fog and cloud computing network. The
research reduces the security risk, energy, delay, and deadline
of applications. The study devises a new dynamic detection
Federated Learning-Based Fraud Detection-Enable Blockchain
System for IoT healthcare applications in Fog-Cloud. The
study divides the learning process into local learning models
at each fog node and then shared to the global master node
to efficiently deal with the entire system. The study formu-
lates this problem as the scheduling problem and makes the
following efforts to solve the considered problem.
•The study devises a novel federated learning aware
fraud-detection enables blockchain IoMT system where
the fraud of data and validation of data and execution
energy and delay being minimized as shown in Figure
1. The system consists of different layers, such as an
application layer consisting of interrelated applications
that share data to process their goals in the network. The
fog-node layer consists of varying fog nodes federated
and fraud blockchain and trains the models at different
nodes to avoid any attack on the storage between various
transactions. The fog-cloud agent (FCA) is the master
node that schedules all requested work based on shared
models to the global federated learning model.
•The research devises the framework-based FL-BETS con-
sisting of different heuristics to solve the problem in
different sub-steps. The heuristics are fraud detection,
blockchain scheme, and task scheduling in the frame-
work. The goal is to process the entire workload of
applications in a different process to validate their quality
of service requirements.
•The study implemented fog nodes near the user applica-
tions to store and train the data to avoid dynamic fraud in
blockchain-enabled with hashing and different primitive.
The local train model shared their training model and the
FCA’s global train model to ensure the system’s overall
performance.
•The scheduler is the dynamic, iterative method that
schedules all workloads’ must be completed with their
deadlines on to the different fog nodes based on their
training model and blockchain requirements. The sched-
uler is non-preemptive, which will not interrupt any
process during execution at the node. Rescheduling will
be possible whenever any failure occurs in the system.
The manuscript has the subsequent sections. Section 2
shows the efforts of existing studies in literature state of the art.
Section 3 shows the problem-solution based on the proposed
architecture, and section 4 shows algorithm implementation
based on heuristics steps. Section 5 determined the perfor-
mances of the proposed schemes with the considered problem.
Section 6 summarizes the effort of the proposed work with the
achievement of the results.
II. EX IS TI NG RE LATE D WOR K OF IOMT SY ST EM
These days, Internet of Medial Things (IoMT) system en-
abled applications are growing progressively to solve different
daily life issues. The most reason is that each traditional
application is connecting to the web via different sensors and
actuators. Whereas many traditional machine learning schemes
supported supervised, unsupervised, and reinforcement learn-
ing are proposed to coach the models of IoT data within the
fog cloud network. Blockchain technology is widely deployed
at distributed fog-cloud networks. However, the traditional
training model of machine learning suffered from overhead,
lateness, accuracy, and lots of factors during the training of
IoT workload within the system.
This work [1] devised the IoMT system based on different
machine learning models for the applications. Recently, feder-
ated learning solved the matter of the normal machine learning
model and trained the various models at local devices, and
shared with the most servers as wiped out for the healthcare
applications [6]. The most goal of this studies to train the
various models of applications at mobile devices during the
processing of healthcare applications. The CNN and KNN
based federated learning aware training models based sug-
gested in studies [2]–[4]. The goal is to attenuate the energy
consumption of devices during training the models locally and
share them with the near fog servers for further manipulations.
However, these studies still didn’t consider the security of
distributed computing.
Security and Privacy-aware federated learning supported
blockchain suggested in [5], [7], [8] for smart-home applica-
tions in the mobile edge cloud and fog computing. The goal is
to scale back the delay of applications and provide security to
the sensitive data of applications during processing outside the
devices. The energy optimization aware blockchain supported
federated learning-based system suggested in [9]–[11]. The
shared and native training models were considered because
the objective with the minimization of energy consumption
of mobile devices and communication networks. However,
existing virtual machine-based systems are still affected by
the delay.
The lightweight IoMT aware decentralized federated learn-
ing for E-healthcare applications suggested in [12]–[20]. The
goal is to attenuate the wait delay, service delay and validate
the info in several nodes. These studies suggested different
machine learning algorithms like deep reinforcement learn-
ing schemes that supported Markov Model and optimized
the multi-objective supported weighting and Pareto optimal.
However, the proposed work is different from the state of art
studies.
The studies [21]–[25] suggested the federated learning-
enabled technology in the fog cloud-aware IoT system was
devised for the different applications. These studies devised a
blockchain-enabled, secure data-sharing architecture for mul-
tiple parties. These studies devised the privacy of data is well-
maintained by sharing the data model instead of revealing
the actual data. Finally, we integrate federated learning into
the consensus process of permissioned blockchains so that the
computing work for consensus can also be used for federated
training. Numerical results derived from real-world datasets
show that the proposed data sharing scheme achieves good
accuracy, high efficiency, and enhanced security. However,
these studies only considered the homogeneous fog and cloud
nodes in their work. These studies considered intelligent
Authorized licensed use limited to: Southeast University. Downloaded on April 09,2022 at 00:17:01 UTC from IEEE Xplore. Restrictions apply.
2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3165945, IEEE Journal of
Biomedical and Health Informatics
3
transport applications and industrial applications with the only
security constraint in their work. However, security, privacy,
and deadlines for healthcare applications are widely ignored
in work.
The study highlights the dynamic security and privacy
challenges and their solutions in the manuscript to the state-
of-the-art studies. The study devises federated learning-based
blockchain-enabled task scheduling (FL-BETS) in the IoMT
system for healthcare applications, which analyzes differ-
ent constraints for optimization, such as validation, delay,
and optimization under deadline for healthcare applications.
The closely related studies [18]–[20] suggested trust and
blockchian enabled for healthcare applications in the dis-
tributed network. However, they did not consider the delays,
fraud, energy, and deadline constraints of the problem inside
the formulation. These studies exploited traditional machine
learning models for training and testing inside the fog and
cloud nodes for healthcare IoT applications. However, these
models have a lot of consumption of resources, energy, and
time and often miss the deadlines for applications.
Federated-Learning Aware Training-Based Privacy Preser-
vation and Fraud-Enabled Blockchain The IoMT System
obtained authentication and authorization issues and threat
attacks. The federated learning approach enables the different
nodes to train and test their privacy and security models
independently and share them with the global model node
for processing. It is the more efficient mechanism used in
the traditional privacy and security mechanisms of machine
learning for IoMT applications.
III. PROP OS ED SO LU TI ON
The study devised the federated learning distributed training
modeling enabled privacy preserving and fraud detection in
blockchain enabled IoMT system for the Internet of Things
(IoT) healthcare applications as shown in Figure 1. The IoMT
system consists of applications, fog nodes, and fog-cloud
nodes with the container microservices. The system considered
the Inumber of IoT applications such as E-Healthcare, E-
Transport, and Smart-Homes. Each application iis coarse-
grained and fully offloads the workload Wito Fog-Cloud
Agent (FCA) for further processing. The number of computing
nodes Kare geographically distributed at different places in
the network. Each node kcan train the data model of all
IoT applications and can share with FCA and other nodes in
the network. The FCA decides on the trained model at local
fog nodes and achieves how to reduce applications’ energy
consumption and latency and execute them under their dead-
lines. The study devises FL-BETS algorithm framework that
consists of different dynamic heuristics as shown in Algorithm
1 This research uses machine learning to implement multiple,
self-improving, and maintainable fraud detection models. We
demonstrate how to train supervised and unsupervised machine
learning models on historical transactions to predict whether
incoming transactions are fraudulent or not.
A. Problem-Formulation of Research
In the research problem, the model consists of Inumber
of IoT healthcare applications, i.e., {i= 1, . . . , I}, where
Fog Cloud Agent (FCA)
ECG
Heartbeat
local Fog Model
E-Hospital
local Fog Model
Blood-
Pressure
local Fog Model
iTM1 iTM2 iTM3
TM1 TM2 TM3
FL-BETS
GTM
i=1 i=2 i=3
B1 B2 B3
K4.B4
k1 k2 k3
Local Fog Layer
IoT Healthcare Applications
Adaptive Training
Adaptive Testing
FL-Fraud Policy
Global Blockchain-Enabled Cloud Computing
Federated Learning Enabled IoMT System
Fig. 1: Federated Learning Aware Blockchain Enabled IoMT
for IoT Healthcare Applications.
{Wi= 1 ∈I}demonstrates their workloads under their
deadlines Di. The proposed model consists of Inumber of
coarse-grained healthcare applications, e.g., {i= 1, . . . iI}.
For executing the workload, the study suggests the hybrid
paradigm enabled fog and cloud nodes that are represented by
{k= 1, . . . , K}with their different speed ζkand resources
r. The blockchain technology with schemes implemented
inside nodes with other blocks, e.g., {B= 1, . . . , N }. Each
blockchain blocks Binside the particular node khas various
attributes such as current block number, current hashing, pre-
vious hashing, transaction data, fraud enabled, and validation
schemes.
1) Security and Privacy Model: The study devised the
symmetric based security model inside blockchain technology
in which encryption and decryption done based on advance
encryption standard (AES). The blockchain technology by
default exploited SHA-256 security algorithm with the public
key and private key during blockchian mechanism during
distributed ledger data in fog cloud network. The study devised
the privacy model and analyzed the fraud detection to avoid
from any privacy issue in the work. The study train and
test the model based on federated learning in the distributed
Authorized licensed use limited to: Southeast University. Downloaded on April 09,2022 at 00:17:01 UTC from IEEE Xplore. Restrictions apply.
2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3165945, IEEE Journal of
Biomedical and Health Informatics
4
TABLE I: Problem Notations
Notations Aims and Definition
ISet of healthcare coarse-grained applications
i ith workload of Iapplications
DiDeadline of the workload ias the hard constriant
WiWorkload of application i
KA set fog and cloud computing-nodes
kThe kth computing-node of K
k−watt Power consumption of node as the soft constraint
ϵkParticular kresource
ζkThe processing speed of node k
N, B A set of blockchains blocks and particular block
T MiTraining and testing model of workload i
xi,k,B Assignment of workload one at a time
blockchain technology.
2) Local Training Model and Blockchain: Initially, the
system will decide offloaded workload from IoMT application
finds particular node or not on the base of following equation.
xWi,k =Assign = 1,
NotAssigned = 0.(1)
Equation (1) determines the either offloaded workload has
been assigned to any particular node or waiting for assignment
in the system. After the workload assignment to any node,
the training and testing analysis sets are divided into training,
testing, and validation in the following way.
T raining =Wi
ζk
× {w= 1, . . . , Wi}.(2)
Equation (3) takes raw data (e.g., audio, video, image, text)
and identify them based on pattern of data and similar pattern
data store in the one cluster.
V alid =T raining ×Wi.(3)
Equation (3) validated the cluster workloads wiof similar
group based on their patterns in the system.
3) Global Model and Blockchain: The study introduces the
local and global validation based on blockchain mechanism in
the study. The goal is to reduce the security risk, delay and
processing cost of applications in the system. Table I) defines
the description of the symbols for the problem.
B. Placement of Fog and Cloud Nodes
The fog nodes are placed very near medical applications,
such as fog nodes located at different hospitals and offering
service at the localization. However, the centralized cloud is
located multiple hops away from user applications and has
high latency but less resource cost. The study considered
the heterogeneous fog cloud nodes as the joint optimization
resources for healthcare applications to keep the resource bal-
ance between fog nodes and cloud. All the fog and cloud nodes
are decentralized they can share data with the blockchian
hashing and make validation based on the proof of work
method.
C. Multi-Constraints Optimization of the Problem
The study considers the different constraints to formulate the
problem, such as data validation, accuracy, resource utilization
which are sets of the delay and processing cost objectives
functions.
D. Processing Delay
Initially, the study calculates the validation delay of data in
the following way.
V alidation −Delay =xWi,k ×V alid. (4)
Equation (4) ensures the validation delay of the single work-
load in the system. The blockchain and execution are calcu-
lated in the following delay.
Execution −Delay =Hashing −time +Wi
ζk
×xWi,k
+F raud −Analysis.
(5)
Equation (5) determines the execution delay of workload i.
Hashing −time =Wi←k, B ←(SH A −256).(6)
Equation (6) determines the encryption and decryption inside
blockchain for workload i.
F raud −Analysis =Wi←k , B ←Data −index. (7)
Equation (7) determines the fraud analysis of workload i.
The power consumption nodes depends upon the blockchain
validation and fraud analysis for each workload to be deter-
mined in the following way. The total delay of workload to
be determined in the following way.
Delay =Execution −Delay +H ashing −time
+F raud −Analysis. (8)
Equation (8) to be determined the all delays of workload i.
P ower −Consumption =W att ×Delay. (9)
Equation (9) determines the power consumption due to all
execution, blockchain and fraud analysis process in the node
for workload. The problem mathematically formulated as
follows.
I
X
i=1
K
X
k=1
N
X
B=1
min Delay , P ower −Consumption. (10)
Subject To
I
X
i=1
K
X
k=1
N
X
B=1
Wi, k, B ≤Di,∀i= 1, . . . I . (11)
All workloads executed under their deadlines as determined
in equation (11).
I
X
i=1
K
X
k=1
N
X
B=1
Wi, k, B ≤ϵk,∀k= 1, . . . K. (12)
All workloads executed under their resource limitation of
nodes as determined in equation (12).
Authorized licensed use limited to: Southeast University. Downloaded on April 09,2022 at 00:17:01 UTC from IEEE Xplore. Restrictions apply.
2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3165945, IEEE Journal of
Biomedical and Health Informatics
5
IV. PROPOSED FL-BETS STRATEG Y FRAMEWORK
In this section, the study shows the processes of FL-
BETS framework which consists of different heuristics as
shown in Algorithm 1. The FL-BETS consists of different
processes and different schemes in the following way. The
study considered the workload assignment problem which is
to be solved in polynomial time that is workload deadline in
the work. Generally, all the workload assignment problem are
NP-Hard problem and to be solved within deadline. Algorithm
Algorithm 1: FL-BETS Algorithm
Input : I , W, T M, K ;
1begin
2for (i=1 to k=1 to K) do
3Call Algorithm 2 order all workload based on
their deadlines;
4Algorithm 2 accepts input at a time instead of
random arrival in the system;
5Sorting i←Wi,∈I;
6Searching and mapping k←ϵk, Di,∈K;
7Call Algorithm 3 to train and test the data on
different local nodes;
8Apply training and testing based on federated
learning at different nodes k, Wi, T M ;
9Call blockchain Algorithm 4 to store secure
data at different nodes with adaptive
validation;
10 End Inner condition
11 End main
1 takes the different constraints as the input, e.g., I, W, T M , K
and process them into different phases. Algorithm 1 is the
framework which consists of different strategies as discussed
in problem description and solve the problem in different steps.
Each strategy to be explained in different sub-sections.
A. Deadline Enabled Sorting and Scheduling
All the workloads have a fixed deadline and are known in
advance in our work. All the workloads are submitted together
in the system at a time. Therefore, all the workloads are
sorted and scheduled based on our work’s lowest deadline
first scheduling method. In this paper, the deadline enabled
scheduler which is the preemptive method that schedules
the workload based on its deadline and privacy and security
constraints. If the workload fails due to privacy or security
issues, the scheduler will reschedule the workload on the
different fog and cloud nodes highlighted in the manuscript.
Algorithm 2 sorts the all workloads based their deadlines.
All the nodes are sorting according to their energy and delay
constraints in network. All the workloads must be executed
under the available resources of nodes until and unless the
assigned workloads finished their executions.
B. FL-Fraud-Detection and security preserving Policy
Adaptive training and testing means training and testing
different models on different fog nodes and sharing them with
Algorithm 2: Delay, Energy and Deadline Efficient
Scheduling
Input : {i=1,. . . , I},{k=1,. . . ,K},
{T M = 1, . . . , T M };
1begin
2Schedule all workloads based on equation (5).
Check the validation resources based on equation
(12);
3Check the consumption of nodes based on equation
(9);
4if (Wi≤ϵk)then
5Apply Schedule-list[i←k←T M ];
the centralized cloud for computation. Based on their training
and testing models, all the trained and tested workloads avoid
fraud attacks and privacy issues. The study devises the feder-
ated learning-enabled fraud detection (FL-Fraud) and privacy
preserving policy in which training and testing progress on
the healthcare workloads by analyzing the fraud detection in
the fog cloud environment as shown in Figure 2. The fog and
cloud nodes are implemented at different layers where local
federated models are trained on the fog nodes and shared to
the global cloud computing for the blockchain-enabled storage
in the system. FL-fraud trains the model to discover fraud
patterns based on proposed adaptive training and testing model
enabled federated learning-enabled policy. The model is self-
learning; it can adapt to new and unexpected fraud patterns.
The study categorized the fraud pattern into two mechanisms:
i=1
k=1
B1
BC
256-SHA Hash
TM1
i=2
k=2
B2
BC
256-SHA Hash
TM2
i=3
k=3
B1
BC
SHA-256 Hash
TM3
Adaptive
Testing
Adaptive
Testing
Adaptive
Testing
Adaptive
Training
Adaptive
Training
Adaptive
Training
k4=TM1, TM2, TM3
Storage
Global Cloud Computing
ECG Heartbeat
i=2
E-Hospital
i=1
Blood Pressure
i=3
Fraud-Analysis
Fraud-Analysis Fraud-Analysis
Fraud-Analysis
FL-Fraud Policy
Fig. 2: Federated Learning Fraud-Detection Analysis and
Policy.
known pattern and unknown pattern. The applications offload
their workloads to the locally available fog nodes for further
processing (e.g., training and testing) before execution to the
global node. The proposed algorithm initially assigned the
particular pattern to each workload and offloaded them to the
respective available fog nodes. The existing proof of work and
proof of stake consensus employ the smart-contract rule to
Authorized licensed use limited to: Southeast University. Downloaded on April 09,2022 at 00:17:01 UTC from IEEE Xplore. Restrictions apply.
2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3165945, IEEE Journal of
Biomedical and Health Informatics
6
practice the fraud on data and validation between nodes in the
blockchain-enabled network. However, these are static rules;
therefore, machine learning-enabled fraud detection based on
random-forest has been implemented in the blockchain to
handle the dynamic fraud in the network. However, central
machine learning fraud-detection algorithms based on random-
forest, artificial neural network, and consensus validation has
delay issue in the centralized training and testing model
for the processing. Therefore, the proposed FL-Fraud policy
delays optimal and accurate handling of known and unknown
fraud in the system. In the algorithm, GT M represents the
Algorithm 3: Adaptive Training and Adaptive Testing
Input : Schedule-list[i= 1 ←k1←T M to
W I ←K←GT M ], F raud[list];
1begin
2foreach (i= 1 ←T M1.k1to I←K.GT M )do
3if (i←T M.k .Matched.Fr aud[list])then
4Training-i←T M.k ←80%.status=Tempt;
5Test-i←TM .k ←20%.status=Tempt;
6else
7i←T M.k .status=Success;
8i←T M.k to GT M ∀k= 1, . . . , K ;
9End-Loop;
global training model based on federated learning, whereas
k.T M is the federated learning model of fog node in the
IoMT system. Algorithm 3 scheme determines any fraud if
there are any chances of tempered data between different
blocks in the blockchain network. Initially, all models train
at other nodes and share the fraud details and the master node
FCA in the system. This way, any dynamic fraud can detect
easily in the system. Algorithm 3 uses machine learning to
implement multiple, self-improving, and maintainable fraud
detection models. Algorithm 3 demonstrates supervised and
unsupervised machine learning models to train the local and
global model on historical transactions to predict whether
incoming transactions are fraudulent or not.
C. Federated Learning Enabled Fraud-Validation of
Blockchain Process and Consensus Method
In this session, the study shows the fraud and validation en-
abled blockchain process and consensus method for workloads
in the system. The existing blockchain consensus methods
widely ignored the delay and energy aspects while evaluating
the intrusion and validation of the nodes in the system. The
study devises the lightweight blockchain consensus, which
is delay and power-efficient and more effective to find the
intrusion and delay with minimum consumption.
The study devises the privacy preserving mechanism in
terms of malware detection in the system. The goal is to safe
data or access control of nodes from malware attacks in the
system. The time complexity of the method to be determined
by nnumber of searching rounds and nof improvement during
execution in the system. Therefore, it is equal to log(n×n)in
Algorithm 4: Federated Learning Enabled Fraud-
Validation and privacy preserving of Blockchain Policy
in Fog-Cloud Paradigms
Input : Schedule-list[i←k←T M ],{B= 1, . . . B},
Data
1begin
2Data=is fraud text to be included in original
workload;
3Pattern= Pattern of the encrypted and decrypted
data;
4Apply Encryption and Decryption based on
equation (6);
5Pattern=k, B, Wi←T M initial pattern of
encrypted workload;
6Pattern’=k, B, Wi←T M offloading from sensor
to local fog node;
7if (Pattern̸=Pattern’) then
8Apply security preserving;
9Determine the fraud based on equation (7);
10 Pattern←Data (list of fraud or unknown
attack);
11 Determine the extra available resource based
on equation (12);
12 Determine the fraud analysis delay based under
the deadline based on equation (8) and
equation (11);
13 Determine the power consumption of fraud
analysis based on equation (9);
14 else
15 Train and validated Pattern on local fog nodes
based on (3) and (3) equations;
16 if (Pattern’̸=Pattern”) then
17 Analyze the fraud between local train models
to cloud storage model based on first given
condition;
the security and privacy method. Algorithm 4 determined the
fraud and validation analysis of workload at different nodes
and trained them locally and shared to the global node. As
shown in Figure 2, the study analyzes the fraud from local
sensor data to local train fog node with the hashing pattern and
fraud pattern. The initial pattern, e.g., pattern’ design during
adaptive local training and testing models (e.g., 80% training
and 20%) in the model. Then further sharing to the remote
cloud, the pattern” is matched on the original data without
affecting any intrusion or attacks. If there is an attack from
a list of intrusion data (e.g., data), the algorithm will recover
them until it gets the original pattern of a particular node.
V. EXPERIMENTAL SET TI NG A ND RE SU LTS
In this paper, the study devised the federated learning-
enabled policy to minimize the energy of fog-cloud nodes
and reduce the delay of applications in the blockchain-enabled
network. Malfeasance systems are constantly rising in an IoT-
enabled fog-cloud setting. Rule-based fraud detection systems
Authorized licensed use limited to: Southeast University. Downloaded on April 09,2022 at 00:17:01 UTC from IEEE Xplore. Restrictions apply.
2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3165945, IEEE Journal of
Biomedical and Health Informatics
7
have traditionally been used to mitigate online fraud, but they
rely on a static collection of rules adaptive and intelligent
experts. This research uses machine learning to implement
multiple, self-improving, and maintainable fraud detection
models. We demonstrate how to train supervised and unsu-
pervised machine learning models on historical transactions to
predict whether incoming transactions are fraudulent or not.
We also explain how to deploy the models to a REST API
to incorporate into an existing business software framework
once trained. This paper widely exploits a demonstration of
the method using an anonymized data transactions dataset.
A. Privacy Preservation Fraud Data Provider Healthcare
Dataset for Experiment
One of the most severe issues facing Medicare is provider
fraud. According to the government, Medicare spending grew
tremendously due to Medicare claim fraud. Healthcare fraud
is an organized crime where peers, providers, physicians,
and beneficiaries, collaborate to file false claims. A thorough
examination of Medicare data has revealed many doctors
who commit fraud. They exploit vague diagnosis codes to
justify the most expensive treatments and drugs. The most
vulnerable institutions affected by these unethical acts are
insurance companies. As a result, insurance companies have
raised their prices, and healthcare is growing more expensive
by the day. Fraud and abuse in the healthcare system can take
various forms. The following are some of the most typical
types of provider fraud: (i) Charging for services that are
never rendered. (ii) Submitting a claim for the same service
twice. (iii) Falsifying information about the service delivered.
(iv) Billing for a more complicated or costly service than was
supplied. (v) Billing for a covered service when the service
was not performed.
This study aims to ”predict possibly fraudulent suppliers”
based on their claims. We will also find significant variables
that will aid in the detection of possibly fraudulent providers’
behavior. Furthermore, we will investigate fraudulent tenden-
cies in provider claims to better predict provider behavior in
the future. The Dataset’s Introduction, We’re looking at in-
patient claims, outpatient claims, and beneficiary information
for each provider for this project. Let’s look at their specifics:
(i) Inpatient Information reveals the claims filed for patients
admitted to hospitals. It also includes information such as their
admission and discharge dates, and they admit d diagnostic
code. (ii) Outpatient Information. This information reveals
the claims filed by patients who visit hospitals but are not
admitted. (iii) Information on the Beneficiary information such
as health conditions, region of residence, and so forth. Table
TABLE II: Nodes Specification of Resources
K ϵj(GB)Core ζj(MIPS) ϵk
k12000000 1 10000 10000PW
k2500000 1 5000 1000PW
k31000 1 1000 500PW
k4100000 4 10000 500000PW
II elaborate the resource specification of system with different
configuration, such as speed, memory, and power consumption
in study. The study exploited stratigraphic statistics tool to get
the graphical results on the generated data in the work.
B. Medical Application Workload
The study gets the three different workload as the coarse-
grained from Kaggle such as ECG heartbeat as images, E-
Heart videos and images and blood pressure as the numeric
text and process them in the system during execution. The
detail of workload shown in Table III. Table III also configured
TABLE III: IoMT Application Workload
i Input Wi(MB) DiFederated-Learning
i=1 Images-Tasks 1000(MB) 1000 (ms) TM1=500 (ms)
i=2 Videos-Tasks 1300(MB) 2000 (ms) TM2=700 (ms)
i=3 Text-Tasks 1028(MB) 2800 (ms) TM3=800 (ms)
the parameter setting of federated learning and its processing
and validation time as the values T M for the simulation in
the system.
C. Federated Learning Training Model
The study suggested the federated learning aware training
model with trained the different dataset of IoT applications
on different fog nodes and shared with the main server fog-
cloud agent (FCA). All nodes can train dataset model of
all applications and easily shared to each other as shown in
the following implementation. %beginlstlisting The source, as
mentioned earlier code defined the blockchain implementation
in IoT systems with different steps. The study implemented the
following baseline approaches in the experimental part, which
are closely similar to the proposed work and already discussed
in the related work part.
•Baseline 1: These machine learning enable methods [1]–
[3], [6] inside blockchain technology for fraud detection
are widely exploited to train the healthcare model and
e-transport and smart-home models. Data mining is used
to classify automatically, cluster, and segment data and
uncover relationships and rules in the data that may in-
dicate intriguing trends, such as fraud tendencies. Expert
systems that encode expertise in the form of rules for
identifying fraud.
•Baseline 2: These dynamic machine learning methods [4],
[5], [7]–[9] are widely exploited to train the healthcare
model and e-transport and smart-home models. The study
implemented all traditional machine learning training
models and implemented blockchain, including schedul-
ing and energy efficiency in fog-cloud methods.
D. Fraud Analysis
Figure 3 shows the performances of the proposed algorithm
on the fraud data in the system. From Figure 3 it can be
observed that, initially, the ratio of fraud and delay is high
such as 60 to 80. After that, the system learns the adaptive
training and testing and improves both the fraud ratio and
delay with the minimum loss as shown in Figure 3.
Authorized licensed use limited to: Southeast University. Downloaded on April 09,2022 at 00:17:01 UTC from IEEE Xplore. Restrictions apply.
2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3165945, IEEE Journal of
Biomedical and Health Informatics
8
Fig. 3: Adaptive Training and Testing Fraud Data.
(a)
600 800 1000 1200
Number of Workloads
0
5
10
15
20
25
Training Delay in Minutes
TMLA
FL-BETS
FTMLA
(b)
200 400 600
Number of Workloads
0.9
0.95
1
1.05
1.1
1.15
1.2
1.25
1.3
Fraud-Detection Delay in Blockchain in Minutes
Baseline1
FL-BETS
Baseline2
Fig. 4: Delay Performance of IoMT Applications With
Different Numbers of Tasks.
E. Result Discussions and Performances of Proposed Schemes
The study conducted the experimental based on different
components such as latency objective with the deadline,
blockchain-enable accuracy and resource leakage performance
of blocks, and energy efficiency of nodes during execution in
the system. Figure 4(a) shows the delay performances while
using the proposed scheme FL-BETS as compared to baseline
1 and baseline 2 during the processing of IoT applications
on the different computing nodes. The main reason is that
all existing studies only focused on resource scalability in-
stead of resource utilization while scheduling workloads on
machines. Another perspective is that the FL-BETS trains the
models at different fog nodes for each IoT workload, with
a minimum training delay instead of a centralized training
model. Therefore, federated learning-enabled distributed fog-
cloud scheduling training optimally compared to a centralized
machine learning training model in terms of delay. Figure
4 (b) the training delay performances of existing training
machine learning algorithm (TMLA) for IoT applications
with workload executions. In contrast, the Fully Training
Machine Learning Algorithm (FTMLA) is widely used to
train the offloaded data on single scalable nodes. However,
these centralized nodes incur long train delays if the number
of IoT applications increases in the system. Therefore, the
FL-BETS distributed training model, e.g., federated learning
model, gained optimal results compared to the existing training
model in terms of delay in the system. Figure 5 (a) shows
the power consumption due to fraud analysis and execution
performances of the scheduling with the training model for
IoMT applications in the system. Figure 5 (b) shows the
power consumption during blockchain validation in nodes
in both figures; the results show that the federated learning
consumes less power for data training as compared to the
existing centralized training model for IoMT applications.
Figure 6 (a) and (b) shows the delay performance during
(a)
200 400 600 800 1000
Number of Workloads
0
5
10
15
20
25
30
35
40
45
50
Power Consumption in Watt
Baseline1
FL-BETS
Baseline2
(b)
200 400 600 800
Number of Workloads
0.32
0.34
0.36
0.38
0.4
0.42
0.44
0.46
0.48
0.5
Blockchain Validity Energy Consumption in Watt
Baseline1
FL-BETS
Baseline2
Fig. 5: Energy Consumption of IoMT During Fraud Analysis
and Execution With Different Numbers of Tasks.
(a)
200 400 600 800
Number of Workloads
0
2
4
6
8
10
12
14
16
Blockchian Delay in Minutes
TBC
FL-BETS
ATBC
(b)
200 400 600 800
Number of Workloads
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
Blockchain-Leakage
Baseline1
FL-BETS
Baseline2
Fig. 6: Blockchain Performances of IoMT Applications With
Anomaly Detection Performances on Different Tasks.
blockchain verification and resource leakage in the distributed
network. The existing traditional blockchain (TBC) algorithm
and active traditional blockchain (ATBC) are widely exploited
in the distributed fog-cloud network. However, they trained
the fraud and security and proof of credibility, sake, and work
cases on single centralized nodes in the peer-to-peer network.
It will face a lot of delay for IoMT applications. Therefore, in
this case, FL-BETS outperforms all existing schemes in terms
of delay in the blockchain process in distributed fog-cloud for
IoMT applications in the system. Besides energy consumption,
delay, blockchain validation, privacy preservation and fraud-
detection performance in the distributed fog-cloud nodes for
IoMT, the deadline is also more critical in IoMT applications
during scheduling in the system. Figure 7 (a) and (b) show
the delay-performances of IoMT applications with the baseline
approaches, e.g., baseline 1 and baseline and proposed scheme
FL-BETS. At the same time, the delay of IoMT applications
with the distributed federated enabled framework FL-BETS
(a)
200 400 600 800
Number of Workloads
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
Deadline Missed Ratio
Baseline1
FL-BETS
Baseline2
(b)
200 400 600 800
Number of Workloads
2.5
3
3.5
4
4.5
5
5.5
6
6.5
7
7.5
8
Deadline Missed Ratio
Baseline1
FL-BETS
Baseline2
Fig. 7: Delay Performance of IoMT Applications With
Different Numbers of Tasks With Anomaly Detection.
Authorized licensed use limited to: Southeast University. Downloaded on April 09,2022 at 00:17:01 UTC from IEEE Xplore. Restrictions apply.
2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3165945, IEEE Journal of
Biomedical and Health Informatics
9
(a)
400 600 800 1000 1200
(a) Application Workloads
0
5
10
15
Delay in Minutes
Baseline1
FL-BETS
Baseline2
(b)
400 600 800 1000 1200
(a) Application Workloads
0
5
10
15
Delay in Minutes
Baseline1
FL-BETS
Baseline2
Fig. 8: Deadline Performance of IoMT Applications With
Different Schemes in Blockchain-Enabled Network.
incurred lower end-to-end delay in the system as compared to
existing schemes. The only reason is that, for each application,
the FL-BETS train model at local nodes is shared to global
nodes for further scheduling. This way, overall delay can be
minimized for IoMT applications. Furthermore, Figure 8 (a)
and (b) show the deadline performances of IoMT applications
besides delay in the system. The results show that FL-BETS
outperforms all existing schemes regarding missing deadlines
and the satisfied deadline for IoMT applications in the system.
F. Finding and Limitation
In the current proposed IoMT system, the study devised
the federated learning-enabled distributed learning process in
a blockchain-enabled IoMT system for different healthcare
workloads. The objective is to minimize privacy and security
issues with the minimum processing time and processing cost
in IoMT work. The study devised the FL-BETS algorithm
framework, which consisted of different methods and achieved
the considered objective with optimal results. However, dy-
namic and run-time unknown attacks are more challenging
for the IoMT and were not considered in this work. Our
subsequent work will consider the adaptive malware-enabled
federated learning method in the blockchain-aware IoMT
system.
VI. CO NC LU SI ON
As shown in the discussion section, the proposed (FL-
BETS) framework minimized energy consumption by 41%
and delay by 28%. The study introduced the mathematical
model. In the performance evaluation, FL-BETS outperforms
all existing machine learning and blockchain mechanisms in
fraud analysis, data validation, energy and delay constraints
for healthcare applications.
The study find the privacy and security issues with the
minimum consumption of the resources in the proposed work
as existing machine learning models did not achieve the
effective utilization of resources in their model and incurred
with the high energy, delay and cost in the system.
In future work, the study will focus on awareness of
mobility fraud and anomaly detection for civil maritime ap-
plications in the blockchain-enabled fog-cloud network. The
cost functions will be determined widely for the system’s
ubiquitous and distributed security constraints. Furthermore,
the proposed system will consider the blockchain-enabled
federated learning methods and the deep-learning decision
model for both preemptive and non-preemptive aspects.
REF ER EN CE S
[1] P. Bhowal, S. Sen, J. H. Yoon, Z. W. Geem, and R. Sarkar, “Choquet
integral and coalition game-based ensemble of deep learning models
for covid-19 screening from chest x-ray images,” IEEE Journal of
Biomedical and Health Informatics, vol. 25, no. 12, pp. 4328–4339,
2021.
[2] Y. Zhou, B. Wang, X. He, S. Cui, and L. Shao, “Dr-gan: Conditional
generative adversarial network for fine-grained lesion synthesis on
diabetic retinopathy images,” IEEE Journal of Biomedical and Health
Informatics, 2020.
[3] Y. Qu, L. Gao, T. H. Luan, Y. Xiang, S. Yu, B. Li, and G. Zheng,
“Decentralized privacy using blockchain-enabled federated learning in
fog computing,” IEEE Internet of Things Journal, vol. 7, no. 6, pp.
5171–5183, 2020.
[4] A. H. Sodhro, N. Zahid, L. Wang, S. Pirbhulal, Y. O. Ouzrout,
A. Sekhari, A. V. L. Neto, A. R. L. De Macedo, and V. H. C.
De Albuquerque, “Towards ml-based energy-efficient mechanism for
6g enabled industrial network in box systems,” IEEE Transactions on
Industrial Informatics, 2020.
[5] R. Saha, S. Misra, and P. K. Deb, “Fogfl: Fog assisted federated learning
for resource-constrained iot devices,” IEEE Internet of Things Journal,
2020.
[6] Z. Lian, W. Wang, and C. Su, “Cofel: Communication-efficient and
optimized federated learning with local differential privacy,” in ICC
2021-IEEE International Conference on Communications. IEEE, 2021,
pp. 1–6.
[7] X. Wang, L. T. Yang, Y. Wang, L. Ren, and M. J. Deen, “Adtt: A highly
efficient distributed tensor-train decomposition method for iiot big data,”
IEEE Transactions on Industrial Informatics, vol. 17, no. 3, pp. 1573–
1582, 2020.
[8] S. Hosseinalipour, C. G. Brinton, V. Aggarwal, H. Dai, and M. Chiang,
“From federated to fog learning: Distributed machine learning over
heterogeneous wireless networks,” IEEE Communications Magazine,
vol. 58, no. 12, pp. 41–47, 2020.
[9] H. Shao, M. Xia, G. Han, Y. Zhang, and J. Wan, “Intelligent fault
diagnosis of rotor-bearing system under varying working conditions with
modified transfer convolutional neural network and thermal images,”
IEEE Transactions on Industrial Informatics, vol. 17, no. 5, pp. 3488–
3496, 2020.
[10] P. K. Sharma, J. H. Park, and K. Cho, “Blockchain and federated
learning-based distributed computing defence framework for sustainable
society,” Sustainable Cities and Society, vol. 59, p. 102220, 2020.
[11] X. Lu, Y. Liao, P. Lio, and P. Hui, “Privacy-preserving asynchronous
federated learning mechanism for edge network computing,” IEEE
Access, vol. 8, pp. 48 970–48 981, 2020.
[12] X. Zhao and C. Huang, “Microservice based computational offloading
framework and cost efficient task scheduling algorithm in heterogeneous
fog cloud network,” IEEE Access, vol. 8, pp. 56 680–56 694, 2020.
[13] A. Lakhan and X. Li, “Transient fault aware application partition-
ing computational offloading algorithm in microservices based mobile
cloudlet networks,” Computing, vol. 102, no. 1, pp. 105–139, 2020.
[14] W. Dou, W. Tang, B. Liu, X. Xu, and Q. Ni, “Blockchain-based mobility-
aware offloading mechanism for fog computing services,” Computer
Communications, vol. 164, pp. 261–273, 2020.
[15] A. Lakhan, M. Ahmad, M. Bilal, A. Jolfaei, and R. M. Mehmood, “Mo-
bility aware blockchain enabled offloading and scheduling in vehicular
fog cloud computing,” IEEE Transactions on Intelligent Transportation
Systems, 2021.
[16] P.-H. C. Chen, Y. Liu, and L. Peng, “Healthcare dataset in machine
learning,” Nature materials, vol. 18, no. 5, pp. 410–414, 2019.
[17] Y. Zhang, Z. Zheng, H.-N. Dai, and D. Svetinovic, “Guest editorial:
Special section on [U+0093]blockchain for industrial internet of
things[U+0094] in ieee transactions on industrial informatics,” IEEE
Transactions on Industrial Informatics, vol. 15, no. 6, pp. 3514–3515,
2019.
[18] S. Farshidi, S. Jansen, S. Espa˜
na, and J. Verkleij, “Decision support
for blockchain platform selection: Three industry case studies,” IEEE
Transactions on Engineering Management, vol. 67, no. 4, pp. 1109–
1128, 2020.
Authorized licensed use limited to: Southeast University. Downloaded on April 09,2022 at 00:17:01 UTC from IEEE Xplore. Restrictions apply.
2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3165945, IEEE Journal of
Biomedical and Health Informatics
10
[19] J. Lockl, V. Schlatt, A. Schweizer, N. Urbach, and N. Harth, “Toward
trust in internet of things ecosystems: Design principles for blockchain-
based iot applications,” IEEE Transactions on Engineering Management,
vol. 67, no. 4, pp. 1256–1270, 2020.
[20] G. Fortino, F. Messina, D. Rosaci, and G. M. Sarn´
e, “Using blockchain
in a reputation-based model for grouping agents in the internet of things,”
IEEE Transactions on Engineering Management, vol. 67, no. 4, pp.
1231–1243, 2019.
[21] Y. Lu, X. Huang, Y. Dai, S. Maharjan, and Y. Zhang, “Blockchain and
federated learning for privacy-preserved data sharing in industrial iot,”
IEEE Transactions on Industrial Informatics, vol. 16, no. 6, pp. 4177–
4186, 2019.
[22] S. R. Pokhrel and J. Choi, “Federated learning with blockchain for au-
tonomous vehicles: Analysis and design challenges,” IEEE Transactions
on Communications, vol. 68, no. 8, pp. 4734–4746, 2020.
[23] U. Majeed and C. S. Hong, “Flchain: Federated learning via mec-enabled
blockchain network,” in 2019 20th Asia-Pacific Network Operations and
Management Symposium (APNOMS). IEEE, 2019, pp. 1–4.
[24] D. C. Nguyen, M. Ding, Q.-V. Pham, P. N. Pathirana, L. B. Le,
A. Seneviratne, J. Li, D. Niyato, and H. V. Poor, “Federated learning
meets blockchain in edge computing: Opportunities and challenges,”
IEEE Internet of Things Journal, 2021.
[25] Y. Lu, X. Huang, K. Zhang, S. Maharjan, and Y. Zhang, “Blockchain
empowered asynchronous federated learning for secure data sharing
in internet of vehicles,” IEEE Transactions on Vehicular Technology,
vol. 69, no. 4, pp. 4298–4311, 2020.
Authorized licensed use limited to: Southeast University. Downloaded on April 09,2022 at 00:17:01 UTC from IEEE Xplore. Restrictions apply.
