Journal of Ambient Intelligence and Humanized Computing (2019) 10:3747–3760

https://doi.org/10.1007/s12652-019-01291-5

ORIGINAL RESEARCH

Fog‑assisted personalized healthcare‑support system for remote

patients with diabetes
Malathi Devarajan1 · V. Subramaniyaswamy1 · V. Vijayakumar2 · Logesh Ravi1

Received: 18 January 2019 / Accepted: 9 April 2019 / Published online: 16 April 2019
© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
Diabetes is featured by the high prevalence and low control resulting in high premature mortality rate. Maintaining the blood
glucose level can bring considerable medical benefits and reduces the risk of diabetes. In real-time, continuous monitoring of
blood glucose level is the major challenge. However, monitoring only glucose level without considering other factors such as
ECG and physical activities can mislead to improper medication. Therefore, the ever-growing requirement for omnipresent
healthcare system has engaged promising technologies such as the Internet of Things and cloud computing. Utilization of
these techniques result with the computational complexity, high latency, and mobility problems. To address the aforesaid
issues, we propose an energy efficient fog-assisted healthcare system to maintain the blood glucose level. The J48Graft
decision tree is used to predict the risk level of diabetes with higher classification accuracy. By deploying fog computing,
an emergency alert is generated immediately for precautionary measures. Experimental results illustrate the improved per-
formance of the proposed system in terms of energy efficiency, prediction accuracy, computational complexity, and latency.

Keywords  Healthcare-support system · Internet of Things · Fog computing · J48Graft classifier · Cloud computing ·
Diabetes

1 Introduction kidney failure (Mathers and Loncar 2006) and referred as

hyperglycaemia. Centers for Disease Control and Prevention
Blood glucose level plays a vital role in maintaining a states that large number of people with the age of 60 and
healthy lifestyle. When the glucose level goes beyond the above is suffered by arrhythmia and people with any age and
normal, it causes severe consequences like heart attack, gender can have diabetes. On the other hand, diabetes cannot
stroke, cardiac arrhythmia, and blindness. When the glu- be cured completely, it can be controlled. The only solution
cose level in blood stream goes down to 60 mg/dl, it causes to reduce the consequences is to maintain the glucose level
cardiac arrhythmia and referred as hypoglycaemia (Robin- in the blood stream by continuous monitoring and regulating
son et al. 2003; Vedanthan et al. 2012). When it rises above the insulin level in real-time.
125 mg/dl, it causes cardiovascular disease, blindness and Internet of Things (IoT) integrates computing devices,
physical objects and people over the network, and enables
* Logesh Ravi them to collect, interact, communicate and exchange infor-
[email protected] mation (Jeon et al. 2009; Rathore et al. 2016b; Sharma et al.
Malathi Devarajan 2017; Jagadeeswari et al. 2018). As web-based services grow
[email protected] rapidly with the development of IoT, users are overloaded
V. Subramaniyaswamy with the information. In recent years, the decision support
[email protected] system has been used extensively to suggest relevant infor-
V. Vijayakumar mation to the user in various fields such as travel and tour-
[email protected] ism, healthcare system, e-commerce, movie, sports (Ravi
and Vairavasundaram 2016; Logesh and Subramaniyaswamy
1
School of Computing, SASTRA Deemed University, 2017; Subramaniyaswamy et al. 2018a; Ravi et al. 2019). In
Thanjavur, Tamilnadu, India
particular, the advancement of the smart devices enhanced
2
School of Computing Science and Engineering, Vellore by IoT has fuelled the development of modern healthcare
Institute of Technology, Chennai, Tamilnadu, India

13
Vol.:(0123456789)

3748 M. Devarajan et al.

systems (Jeon et al. 2016; Wu et al. 2016; Shi et al. 2017; terminal node and the cloud server, saves bandwidth and
Arunkumar et al. 2019; Chehri et al. 2010). Health Internet thereby reduces the latency. It also facilitates fault detection,
of Things (HIoT) connects wearable sensors with the smart interoperability, and mobility (Gia et al. 2018). In this arti-
device to sense, collect and transmit data through the net- cle, a fog-assisted healthcare system for remote patients with
work continuously (Dziak et al. 2016, 2017; Wu et al. 2017, diabetes is proposed. To provide improved classification
Rathore et al. 2017). But, it could be a challenging task for accuracy (Logesh et al. 2018b, 2019), the system monitors
a service provider to manage the tremendous amount of not only physiological conditions but also considers contex-
information generated by IoT (Vairavasundaram et al. 2015). tual information (Wang et al. 2018; Devarajan et al. 2019).
However, due to lack of power supply, memory constraint, The rationale behind the proposed fog-assisted health-
computational time, communication cost and efficient man- care support system is to identify risk level of diabetes and
agement of sensors and mobile devices (Babar et al. 2018), generate an emergency alert with an intention to reduce the
high-performance computing facility and massive storage premature mortality. As the acquired physiological health
infrastructure are in demand. In this scenario, the promising conditions are processed and analyzed at the edge where
potential of cloud computing has been widely used to ensure the data are originally generated, we can save more lives
the quality of the healthcare system. instantly. It significantly reduces the storage space and net-
In today’s world of IoT-based healthcare system, cloud work bandwidth and latency. The key contributions of the
server serves as the backbone for the health data repository. work are as follows:
Cloud computing is defined as the centralized data center,
where various computing resources provide services per • Design and development of a fog-based computing model
pay over the internet (Ghanavati et al. 2017). Healthcare to monitor, predict and control the risk of remote diabetic
service providers deploy the services on the cloud to utilize patients in real-time based on physiological conditions.
the maximum resource, thereby reduces the processing time • Risk assessment of diabetes patients at regular intervals
and cost. Considerably it increases the number of users and to examine any change in glucose level to generate an
service requests. And the personalized suggestion of food alert message and provides precautionary measures in
and drugs based on users’ current situation is a major con- an emergency situation.
cern (Subramaniyaswamy et al. 2017b; Logesh et al. 2018a; • Detailed analysis of real-time health information and
Logesh and Subramaniyaswamy 2019). However, the central physical activities in different circumstances to evaluate
storage, pre-processing, organization, and management of the performance of the proposed fog-assisted healthcare
medical data on the cloud are highly vulnerable to several system.
data threats and privacy attacks as it carries users’ sensi-
tive information (Sajid and Abbas 2016; Yaqoob et al. 2017; The remainder of the article is organized as follows:
Subramaniyaswamy et al. 2017a). To address the above- Sect. 2 describes the literature related to healthcare services
mentioned issues, anonymous data processing is essential. and fog computing in the field of healthcare support sys-
To take advantage of cloud computing, an efficient way is tem. The preliminaries are presented in Sect. 3 for a clear
to hide the real identities of the user (Sarkar et al. 2017). In understanding of the work. Section 4 highlights the services
addition, it also poses problems like high latency, network provided by the fog computing paradigm in the healthcare
traffic, mobility and reliability (Devarajan and Ravi 2018; system. In Sect. 5, the main components of the proposed
Wang et al. 2017; Jangiti et al. 2019). Employing fog com- fog-assisted healthcare system are presented. Section 6 pre-
puting platform as an intermediary between the cloud server sents the case study on diabetes with J48Graft classifier and
and the end user becomes an inevitable trend to analyse and alert generation. Experimental evaluations and results are
manage the big data (Bonomi et al. 2012; Ahmad et al. 2016; discussed in Sect. 7. Finally, Sect. 8 summarizes the conclu-
Sankar et al. 2019; Pinjari et al. 2018). sions and future works.
Better and timelier healthcare service is provided by real-
time monitoring of users’ physiological conditions. For real-
time processing of health data, HIoT are utilized to gener- 2 Related work
ate an enormous amount of health-related information to
examine the situation (Ahmad et al. 2018). It significantly With the help of HIoT, physiological condition of the user is
increases the network traffic, computational complexity, monitored and transmitted to the medical server periodically.
and latency. Therefore, to optimize the healthcare system, Any delay in transmission or missing vital signs may lead
an alternative paradigm called fog computing is used as a to serious consequences. Hence, many efforts were made
gateway between the cloud server and the terminal device to transmit the health data effectively to the remote server
(Natarajan et al. 2019; Asaithambi et al. 2019). Fog comput- to improve the precision of diagnosis. Emerging technolo-
ing effectively minimizes the communication between the gies such as wireless technology, smart sensors, body area

13
Fog‑assisted personalized healthcare‑support system for remote patients with diabetes 3749

network, Internet of Things, cloud server and medical big network traffic, computational complexity, and delay trans-
data have been widely used in the modern healthcare system mission. To optimize the process, a fog computing paradigm
to minimize the processing cost and time; significantly it acts as a gateway between the end user and cloud server.
increases the wellness of remote patients. IoT is considered Bonomi et al. (2014) present the features and properties of
as a revolution due to its ability to interconnect and commu- an intermediate fog computing paradigm for critical IoT
nicate with other devices over the internet. Many research- applications. Yannuzzi et al. (2014) examines the complex
ers proposed IoT based models to monitor patients’ health security challenges of cloud computing and presents fog
condition. computing as a most promising technique to compact with
Kiran et al. (2014) proposed an IoT enabled data acqui- massive storage and computing to support IoT based applica-
sition and transmission system based on an adaptive rule tions. A case study presented by Gia et al. (2015) highlights
engine on ECG data. Performance evaluation shows that the enhancement of health monitoring system by exploiting
the proposed scheme consumes less power and reduces fog paradigm at the edge on ECG feature extraction. As most
network traffic. Yang et al. (2014) proposed a framework of the symptoms are analogous to mosquito-borne diseases,
called intelligent home (iHome) which integrates intelligent accurately classifying the diseases helps to medicate the
medicine box (iMedBox), pharmaceutical packaging (iMed- patient properly. Thus Vijayakumar et al. (2018) proposed an
Pack) and biomedical sensors (Bio-Patch) with telemedicine intelligent system for an effective detection and classification
to enhance the healthcare service. Rathore et al. (2016a) of mosquito-borne diseases. Thus to improve the efficiency
designed a complete scheme for creating a smart city using of the healthcare system, we propose the fog-assisted remote
the big data generated by IoT. The proposed scheme con- patient healthcare system for diabetes.
sists of four tiers for data collection, aggregation, commu-
nication and decision making. Subramaniyaswamy et al.
(2018b) developed an ontology-based food recommenda- 3 Preliminaries
tion model for ProTrip and evaluated it on the real-time IoT
based healthcare system. Selvan et al. (2019) developed an In this section, the preliminaries and the technical back-
ontology-based recommendation model in association with ground essential for a clear understanding of the proposed
the fuzzy theory to reduce the manual effort for prescription fog-assisted healthcare system is presented. The J48Graft
and recommendation of food and drugs for chronic patients. classifier is employed in the healthcare system to effec-
In recent years, as a promising emerging technology, tively classify and discover the risk level of diabetes. So
cloud computing provides massive storage, computing, and that remote patient can maintain their blood glucose level
resource sharing facility at low cost. Luo and Ren (2016) and can be able to prevent premature mortality.
studied and analyzed the application of cloud computing
on medical data monitoring and management. Simulation 3.1 Fog computing
results illustrate that the proposed architecture improves
efficiency by 50%. Abawajy and Hassan (2017) present a With the rapid development of wireless technology, smart
pervasive patient health monitoring (PPHM) scheme by devices, and customized healthcare applications, IoT gen-
integrating IoT and cloud computing for real-time moni- erates an enormous amount of medical data. Such medical
toring of ECG information for congestive heart breakdown big data can be of different form like text, multimedia, and
patients. Experimental results demonstrate that the proposed image that needs to store, analyse and process at the cloud
PPHM facilitates energy efficiency, scalability, and flexibil- server. Handling medical big data at the cloud leads to high
ity. Hassan et al. (2017) proposed a hybrid network model latency, network traffic, and security issues. Hence, it requires
that integrates both cloud and Wireless Body Area Network a new computing platform called fog computing to minimize
(WBAN) for efficient medical media data sharing. An adap- the burden of the cloud. The fog computing layer acts as a
tive streaming technique was used to reduce packet loss, and flyover between the terminal device and the traditional cloud
simulation results show the feasibility of the model in terms server to bring the cloud services close to the network edge,
of reduced delay and the real-time application of medical enabling refined and secured healthcare services. The ideal
data transmission. Malathi et al. (2019) proposed a disease place to examine real-time health data is the network edge,
prediction support system in combination with fuzzy set where data are originally generated. The key features and
theory, case-based reasoning, and KNN. They also extended advantages of the fog layer is distributed computing, decen-
the system to preserve the privacy of the user information tralized storage, local data analytics, data pre-processing, data
through Paillier Homomorphic Encryption. trimming, temporary storage, data security, and privacy. Many
As more and more healthcare service provider would IoT-based applications like health monitoring system demands
like to make use of a centralized cloud server for data stor- both distributed fog and a centralized cloud server to perform
age analytics, it significantly raises the issues like latency, big data analytics efficiently. By using fog computing as an

13

3750 M. Devarajan et al.

intermediary in the real-time healthcare system, it can harvest 4 Services provided by the fog layer
the following benefits:
Fog computing paradigm facilitates local storage, compu-
• Sensors are energy constraint device, thus by shifting tation, networking, and other IoT-based application-spe-
energy consuming services and computation to fog layer cific services. The health monitoring system has a distinct
provides energy efficiency. set of desires, for instance, reliability, remote monitoring,
• It is necessary to analyse and work on the situation imme- real-time data processing, security, and privacy. Figure 1
diately (< second) to avoid serious consequences. highlights the general services provided by the Fog layer
• Data collection, pre-processing, analysis and decision mak- and each service are explained in detail in the following
ing takes place at the edge, thus reduces cloud computing subsections.
cumbersome.
• Millions and billions of Internet of Things are willing to
generate huge data across the world. Hence to reduce net- 4.1 Real‑time medical data management
work traffic and bandwidth, fog computing plays a vital
role. Real-time medical data collected from various sensors are
• By transmitting reduced, processed and compressed data processed locally to extract featured information for analy-
to the cloud facilitates low latency. sis and feedback generation. According to the framework
• Reliable data transfer can be ensured by providing security proposed, Fog layer continuously receives the sensed data,
features for data privacy and intrusion protection. thus data management in this layer is crucial to provide
a fast response regarding the situation and notification in
3.2 J48Graft classifier case of emergency. In the healthcare system, data manage-
ment becomes more trivial since latency and ambiguity
The decision tree is considered as an important candidate for might cause undesirable scenario. The operating system
classification and decision-making process due to its ability to of the smart device used in the fog has a database for
handle a dataset with a high error rate. In k-nearest neighbour local storage and data can be stored in compressed for-
(Maillo et al. 2017; Subramaniyaswamy and Logesh 2017), mat. Other functionalities such as pre-processing, analy-
classification is achieved based on similarity measurement, sis, feature extraction and encryption are also carried out.
and it performs well only for smaller datasets. Random tree After receiving the sensory information, data filtering is
(Kalmegh 2015) builds a decision tree with k-random charac- the first and foremost task of a data processing unit to
teristics, but it falls easily to overfitting. Multilayer Perceptron remove noise and distortions. The lossless compression
(Adeyemo et al. 2015) is a commonly used neural network technique is then applied to the filtered data to reduce
algorithm which is used to classify non-linearly separable communication overhead and computational complexity
datasets. J48Graft (Hayashi et al. 2016; Panigrahi and Borah of sensors. To obtain robust and efficient information, sen-
2018) generates the enhanced classification results with a min- sory data are fused by removing redundant and irrelevant
imized error rate for multiclass datasets. The key advantage data. Consequently, data reduction provides local data
of using the J48Graft classifier is to improve the probability analysis, thus improves the performance of the healthcare
of the classification accuracy for the raw data and to mini- system by reducing latency, bandwidth and unnecessary
mize the error rate. It is preferred because of its less execution data communication to the data centre. In addition, local
time on the multiclass dataset. In the proposed fog-assisted storage space also acts as a cache memory to facilitate the
healthcare system, J48Graft decision tree classifier is used to continuous flow of data.
discover the risk level of a diabetic patient as hypoglycaemia,
normal, pre-diabetes and hyperglycaemia by considering the
influential attributes such as blood glucose level, body weight, 4.2 Smart fog device management
physical activities and diet. By deploying J48Graft classifier,
the proposed system predicts the risk level of diabetes with Health-IoT consists of resource constraint devices, and it
enhanced accuracy in real-time to prevent the user from pre- demands proper management. Fog device executes com-
mature mortality. plex algorithms and computations to extract features. It
performs clinical diagnosis using information collected
from wearable sensors and provides relevant feedback and
reports. Hence it consumes more energy, and it is neces-
sary to manage the fog device properly. It needs to charge
the battery associated device once in a day. Mobility is

13
Fog‑assisted personalized healthcare‑support system for remote patients with diabetes 3751

Fig. 1  General services provided by the fog layer

another concern in order to avoid service interruptions 4.4 Event management and personalized

and packet loss. In case a patient is moving from one loca- healthcare
tion to another by wearing medical sensors, it should not
affect the monitoring process at the new location. Device Real-time remote patient health monitoring system experi-
discovery helps to recognize the new device entering the ence several events like patient health conditions, physical
domain and offers uninterrupted service. activities, and environment. When a serious event occurs,
the event management ensures automatic medical service
and also notifies the experts, family members and even
4.3 Energy efficiency and latency patient about the situation. According to the patients’ health
conditions and requirements, the real-time health monitoring
Sensors used in the healthcare system are resource con- system is required to be configured for various healthcare
straint, thus it must be designed for a particular pur- applications. Personalized healthcare is provided by defin-
pose not for generic tasks to avoid consumption of more ing a system parameter based on the priority and health
energy. It significantly reduces the overheads of data pro- condition. Adaptive rule-based approaches and complex
cessing at the sensor level. Similarly, latency is another machine learning algorithms are used to facilitate personal-
critical requirement because any delay in disease diagno- ized healthcare service. For instance, if abnormal cardiovas-
sis and decision making might cost a human life. Moni- cular related information is detected during monitoring, the
tored sensory data is stored and processed locally in the system should be capable of learning the priority of the situ-
fog layer, from where imminent notification is sent to ation and adaptively adjust the heart associated parameters.
experts and patients in an adverse situation. Data filter-
ing method can be used to remove noise and redundant 4.5 Privacy and security
data to reduce transmission latency. When the internet
connection is unavailable, the smart fog device should In common, security and privacy is crucial for every
re-transmit the unsent or lost packet once the connection real-time healthcare application because a single leak-
is re-established. age or integrity of the health parameter might cost severe

13

3752 M. Devarajan et al.

consequences. If any of the Health-IoT devices are hacked, intermediate computing platform called fog is introduced
the complete health monitoring system can be inhibited by between the cloud and the end devices which complement
an intruder. Therefore, robust cryptography technique should cloud assisted Health-IoT by additional services. The
be applied at both sensor level and fog layer for confiden- proposed fog-assisted healthcare system is illustrated in
tially, reliability and integrity. In addition, each sensor in the Fig. 2. It consists of data collection, fog computing, alert
network should prove their legitimacy with the smart device generation, centralized storage, and data analytics.
and key management service is utilized to facilitate secure In the propose fog-assisted healthcare system, users’
data transmission. Moreover, as real-time monitoring system various physiological conditions such as physical activi-
consists of patients’ sensitive information and usually be ties, glucose level, blood pressure, and body temperature
transmitted via unsecure channel, it is necessary to prevent are recorded by wearable sensors. Contextual informa-
unauthorized access by intruders. tion like temperature, time, location, humidity can also
be supplemented with the health data to predict unusual
patterns of the situation more precisely. Then the sensed
5 Fog‑assisted healthcare support system raw data from various sensors are transmitted to the fog
layer periodically by means of smart devices through Blue-
The implementation of IoT in the healthcare system is tooth, Wi-Fi or ZigBee. The functions like anonymous
expected to connect millions of resources, power and encryption, aggregation, feature selection, classification
bandwidth constrained devices into the network and are performed in the intermediate fog layer. The network
exchange data over the internet. Medical sensors are typi- layer is used to transfer processed health data from lower
cal examples of such devices that continuously monitor fog layer to upper cloud layer securely through various
health-related information, but are not capable of storing networks. And in the cloud computing layer, data storage,
and processing the data. Hence cloud computing has been analytics, statistical research, broadcasting, patient-centric
introduced as central storage for medical big data pro- recommendations are performed. Finally, the service layer
cessing. The large scale usage of cloud computing signifi- provides the user-friendly web application for feedback,
cantly raises the latency and network traffic. Therefore, an report visualization, and suggestions. The key components

Fig. 2  General overview of the fog-assisted healthcare system

13
Fog‑assisted personalized healthcare‑support system for remote patients with diabetes 3753

of the proposed fog-assisted healthcare system are elabo- can process the data locally, analyze and predict unusual pat-
rated in the subsequent sections. terns. In case it monitors that the vital parameter exceeds the
defined threshold level, it instantly generates an emergency
5.1 Perception layer alert to the medical experts and family members through a
smart device. Thereby, it significantly reduces the serious
Perception layer comprises of functions like data sensing consequences like death by providing timely services.
and acquisition. Sensors attached to the human body collects
vital physiological information such as pulse rate, insulin
5.3 Network and communication layer
level, blood pressure, body temperature, sweating, physical
activities and contextual information like humidity, air, room
Network and communication layer provides end to end con-
temperature periodically. Contextual information helps in
nectivity between IoT involved in the healthcare system. It
identifying unusual patterns and makes more precise health-
acts as a bridge for data transmission between the edge and
care service. Then the sensed information is transmitted to
the cloud layer. Like in the traditional OSI model, Network
the upper layer for local processing. To make the system to
layer collects the medical data stably from fog platform and
be energy efficient, co-ordinator/hub is used to collect and
transmit them to the upper cloud server safely. The com-
aggregate the heterogeneous data monitored by various sen-
munication layer chiefly consists of smartphone and other
sors through a wireless channel. Depending upon the situa-
local networks. Smartphone that acts as a co-ordinator is
tion, devices in health-IoT dynamically or statically organ-
utilized to collect sensory data and encapsulate aggregated
ized in a tree or mesh structure. Such topology is suitable for
sensory information into single packet in a proper format,
the patients in a smart home or smart hospital. In the case of
then made ready for transmission to cloud through Wi-Fi
outdoor patients, we require star topology where each sensor
or other types of networks. Utilizing various network and
communicates directly with each other and the co-ordinator.
smart device provides flexible and scalable transmission and
Since data sensing sensors are subject to energy constraint
communication. Robust routing protocol and MAC protocol
device, smart device is used as an application interface. Each
can be widely employed to deliver the data securely and
sensor needs to prove their legitimacy with the co-ordinator
provide proficient communication between devices. Since
to avoid security threats like eavesdropping and man in the
fog computing is used in the proposed healthcare system, it
middle attack. After the successful verification, co-ordinator
ignores unnecessary communication resulting in low latency
encapsulates the heterogeneous data in a specific format and
and network traffic.
adds metadata in the header to identify the patient and smart
device. The packet is then forwarded to the upper fog layer
through Bluetooth or ZigBee. 5.4 Cloud computing layer

5.2 Fog computing layer Cloud layer in the IoT-based healthcare system facilitates
centralized storage, large scale computation, data manage-
Fog layer is the premier unit of the proposed remote patient ment, and data analytics. The health information transmit-
healthcare system. As the number of user increases with ted by fog layer through various networks are stored and
the development of wireless technology, smart devices, and managed in the cloud server. An application interface is
customized healthcare applications, HIoT generates an enor- developed for caregivers to dispatch alerts or notification,
mous amount of medical data. The fog computing layer acts reports, feedback, and other health related recommenda-
as a flyover between the terminal device and the traditional tions. The authorized consumers like general practitioners,
cloud server to bring the cloud services close to the network specialists, research institutes, emergency service have an
edge, enabling refined and secured healthcare services. The ability to log in and put up their request by providing patient
ideal place to examine real-time health data is the network information. After verifying the legitimacy of the consumer
edge, where data are originally generated. The key features and requirement of the request, consumers are allowed to
and advantages of the fog layer are distributed computing, access the health information. In short, the integration of fog
decentralized storage, local data analytics, data pre-process- computing paradigm with the cloud-assisted health-IoT pro-
ing, data trimming, temporary storage, data security, and vides intelligent healthcare services. The cloud server also
privacy. As fog provides localization, it facilitates location provides reports and notification to patients and their family
awareness, local computation, low latency, and local stor- members via the mobile application. Though fog comput-
age; whereas the cloud server provides global centraliza- ing is capable of providing advanced healthcare services,
tion. The controller/hub in the fog layer gathered the sensed limited storage capacity, accessibility of fully updated health
data and transmits the processed health information to the record, running complex algorithms, responding queries
cloud server for global storage and management. The fog with requested data can be addressed only by cloud. Thus

13

3754 M. Devarajan et al.

the integration of cloud and fog platform should be used to data into a single packet by extracting required information
implement effective healthcare services. and filters out unnecessary and redundant information to
generate a sequence file. It compares each sensory value
5.5 Application or service layer with the defined threshold value and stores them in a local
database. In case any abnormal value is detected, immedi-
Application layer facilitates end-user connectivity between ately fog device generates an alert message and notifies the
care centres, hospitals, research institutes, emergency ser- concern people about the condition. If it is normal, it simply
vice, police stations, and diagnosing labs. It offers various stores in the database for future analysis. Finally, aggregated
healthcare services such as chronic disease monitoring, post- information is moved to the cloud server for data analyt-
operative cases, physical activity recommendations, and ics, classification, disease prediction, and prevention, when
emergency services. Application developers can make use it reaches the threshold value, time limit or maximum file
of the application programming interface to develop service- size. The cloud server first verifies the legitimacy of the fog
related applications. A mobile application is also developed device and the report is stored for deep analysis. The health
for supporting the healthcare system by providing notifica- record of the corresponding user is updated habitually and
tion service and feedback. When the cloud server receives maintained securely in the cloud server.
the alert message or notification from the monitoring system,
it immediately generates the message to an end users’ mobile 6.1 Categorization of blood glucose level
phone to enlighten about the emergency situation. Health-
care service providers and users can openly take advantage Based on the pre-defined threshold value, the glucose level
of cloud resources through smart devices. Finally, it provides in the blood stream is categorized into four levels such as
services to the user as a graphical user interface for feedback hypoglycaemia or low glucose level, normal, pre-diabetes
and visualization. and hyperglycaemia or high glucose level. If the fasting
blood glucose level (FBGL) is below 60 mg/dl, then it is
called hypoglycaemia. If the value ranges between 60 and
6 Monitoring diabetic patients: a case study 100 mg/dl, then it is normal. In pre-diabetes, it ranges from
100 to 125 mg/dl. When the level goes above the 125 mg/dl
The real-time applicability of the proposed fog-assisted are diabetes. Usually, the fasting blood glucose level is tested
healthcare system is explained in this section. The main aim before breakfast. Similarly, testing glucose level before and
is to show the effectiveness of the major components of the after 2 h of eating is called the oral glucose tolerance test
proposed system. In this work, we concentrate on monitor- (OGL). In this case, glucose level below 140 mg/dl is nor-
ing the users to diagnose the risk level of diabetes. Let us mal, ranges between 140 and 200 mg/dl is pre-diabetes and
consider that there is a large number of registered users; above 200 mg/dl is referred to diabetes according to Ameri-
each user is assigned a unique identity for documentation, can Diabetes Association.
report generation, and other purposes. The user is required to
maintain their profile with personal information along with 6.2 User classification based on J48Graft classifier
current health condition, treatment history and medication
took so far to generate precise healthcare service. Users’ pro- The blood glucose level of the user varies due to the physical
file contains sensitive information, and they are susceptible activities, diet, habitat and environmental conditions. Thus,
to data threat and privacy attacks. Therefore, to ensure the a learning model is needed to predict the risk level of users
privacy of the user, the real identity of the user is masked by to avoid serious consequences like death. J48Graft generates
pseudo-identity. Based on the pseudo-identity, users’ profile the enhanced classification results with a minimized error
is indexed on the cloud server. The purpose of making the rate for multiclass datasets. The key advantage of using the
fog to be blind is to preserve the users’ private data. Thereby J48Graft classifier is its less execution time on the multi-
making the identity of all patients into blind, at any time the class dataset. Decision tree grafting is used to improve the
information can be fed into the cloud through the fog. classification results, thus it avoids generating the complex
A registered user is assumed to be comprised of the trees. Grafting is applied to the decision tree to reclassify
number of sensors, each with a unique identifier to monitor the instances which consist of misclassified data or no train-
heterogeneous physiological parameter. And also a smart ing data. It significantly reduces the prediction error. In the
device with GPS (global positioning system) service to proposed fog-assisted healthcare system, J48Graft decision
locate the user. A sensor fixed to the patients’ body con- tree classifier is used to discover the risk level of the diabetic
verts the monitored information into a mobile readable patient by considering the influential attributes such as blood
data and sends them to the smart device through ZigBee glucose level, body weight, physical activities, and diet. By
or Bluetooth. Fog device encapsulates the obtained health deploying J48Graft classifier, the proposed system predicts

13
Fog‑assisted personalized healthcare‑support system for remote patients with diabetes 3755

the risk level of diabetes with enhanced accuracy in real- range between 25 and 28 kg/m2 was used. Each subject has
time to prevent the user from premature mortality. to follow the set of rules which contains 18 different physical
exercises such as cycling, lying, ironing, standing, vacuum-
6.3 Alert generation ing, walking upstairs, walking downstairs, normal and Nor-
dic walking, sitting, running, watching TV, computer work,
Alert generation is considered as an important component playing soccer, car driving, rope jumping, folding laundry
of the proposed fog-assisted healthcare system as it gener- and house cleaning. The amount of calories consumed after
ates notification about the current situation and risk level to each activity is calculated for analysis. Altogether, more than
the users’ smartphone. It also stores all the information on 10 h of activity were monitored and presented for experi-
the cloud server about the alert generation such as physical mental study.
activity, location, time, diet, habit and contextual informa-
tion responsible for the adverse situation. These activities 7.2 Efficiency of J48Graft classifier
are monitored automatically by the wearable sensors or
smart device for an equal interval of period. The pre-defined To identify the risk level of diabetes such as hypoglycaemia,
threshold values are a vital parameter for the efficient health- normal, pre-diabetes or hyperglycaemia, J48Graft classifier
care system. If the glucose level goes beyond the threshold is utilized. An experiment was conducted on diabetes dataset
value in the blood stream, an alert is generated to notify the to predict the risk level of the user, and the J48Graft classi-
user about the critical situation. If the user is in a normal fier is implemented in Weka 3.6. To evaluate the efficiency
stage, then the system recommends the precautionary steps of J48Graft classifier, several statistical measures were per-
to prevent the prevalence of diabetes. If the user is at a risk formed. The higher success rate and the lower failure rate
level, then the system recommends proper medication to the indicate the enhanced performance of J48Graft classifier.
user. The algorithm used to generate alert messages is given The success rate is defined as the percentage of correctly
in the Algorithm 1. classified users to the total number of users, whereas the
failure rate is the percentage of inaccurately classified users
to the total number of users. The evaluation metrics such as
7 Experimental Evaluation precision, recall, sensitivity, specificity, and accuracy are
used to show the correctness and completeness of classifiers.
The performance efficiency of the fog-assisted healthcare Analysis of such statistical measure shows that the proposed
system in real-time scenarios is evaluated in this section. system is capable enough to identify the risk level of diabe-
Initially, the dataset is collected and the significant health tes with high accuracy. If the classifier is capable enough to
parameters are analyzed to predict the risk level of a user. provide high values for recall and accuracy, then it automati-
Then the user is classified as either diabetic or not to gener- cally generates high precision value. Higher the precision
ate treatment recommendations as per the situation. value means that the classifier can minimize the error rate.
The J48Graft classifier yields high recall value, thus the
7.1 Dataset description statistical measures display the proposed fog-assisted health-
care support system with J48Graft classifier is proficient
The dataset for blood glucose level and physical activities enough to categorize the diabetic patient with enhanced
are not available in the repository. To verify the efficiency classification accuracy. Higher the sensitivity and lower the
of the proposed healthcare system, the two different datasets specificity value leads to less classification errors. Hence,
from the UCI repository were utilized. For our study, we to portray the improved performance of J48Graft classifier,
consider the UCI diabetes dataset (https​://archi​ve.ics.uci.
edu/ml/datas​ets/Diabe​tes) that contains 70 sets of data on
diabetic patients’ blood glucose level recorded for several
weeks in different time periods. The blood glucose level
is monitored periodically for analysis at normal pre and
post breakfast, snacks, lunch and supper. The main aim
is to sustain the blood glucose level by diet and physical
activities. Hence, we used the UCI PAMAP2_dataset (http://
archi​ve.ics.uci.edu/ml/datas​ets/PAMAP​2+Physi​cal+Activ​
ity+Monito​ ring)​ for the experiments and analysis of diabetic
patients. The PAMAP2: Physical Activity Monitoring Data-
set contains nine subjects i.e. one female and eight male, Fig. 3  Comparison of different classification algorithms in terms of
aged between 27 and 32 years and BMI (body mass index) evaluation metrics

13

3756 M. Devarajan et al.

Table 1  Diabetic patients’ blood glucose level

Date Time Code Value

09-01-1990 16:48 62 226

09-01-1990 16:51 33 1
09-01-1990 16:51 34 8
09-01-1990 22:25 64 154
09-02-1990 08:01 58 66
09-02-1990 08:03 33 2
09-02-1990 08:03 34 10
Fig. 4  Accuracy comparison of different classification algorithms 09-02-1990 11:40 60 136
09-02-1990 11:42 67 0
three different algorithms such as K-nearest neighbour, ran- 09-02-1990 17:15 56 330
dom tree, and Multilayer perceptron are used for experimen- 09-02-1990 17:17 62 319
tal evaluation. Figure 3 depicts the comparison of different 09-02-1990 17:18 67 0
algorithms with J48Graft classifier in terms of statistical 09-02-1990 17:18 33 2
evaluation metrics. The comparative chart of evaluation 09-02-1990 17:18 34 8
metrics shows that the J48Graft decision tree classifies the 09-02-1990 21:57 64 227
risk level of diabetes more efficiently than other classifica- 09-03-1990 03:21 56 92
tion algorithms. Figure 4 presents the comparison chart of 09-03-1990 03:23 56 49
different algorithms in terms of classification accuracy. It 09-03-1990 08:04 58 192
shows that the J48Graft based classification perform better 09-03-1990 08:06 33 3
than other classification algorithms on multiclass datasets. 09-03-1990 08:06 34 9
09-03-1990 12:38 60 115
7.3 Experimental results and discussions 09-03-1990 17:05 56 288
09-03-1990 17:06 62 287
The fog-assisted remote patient healthcare system is evalu- 09-03-1990 17:07 33 1
ated on a smartphone with Snapdragon 410 Quad-Core
clocked at 450 MHz with 2 GB of memory. The sensory
information measured by wearable sensors for a particular pre-prandial blood glucose level measurements of a patient.
interval of time has been used for activity monitoring. The While analyzing the pre-prandial records, we found that
health information generated by sensors is forwarded to a fog most of the pre-dinner readings crossed the higher limit,
device, where they are aggregated and encapsulated into a but infrequently before the breakfast. Therefore, the patient
single packet, appended with header information, and then is suggested to burn more calories before dinner and change
transmitted to the cloud server for storage. Subsequently, the food habit at dinner thereby increase the production of
based on the readings and calories consumed during physi- insulin by physical activity.
cal activity, medical experts suggest recommendations to Patients with diabetes can either be low or zero produc-
enhance the wellness of a patient. After collecting all the raw tion of insulin by the beta islet cells of the pancreas. The
data, it is subjected to pre-processing for removing redun- main aim is to fall the blood glucose level close to normal
dant data and filling missing values. Then the intermediate range by diet and physical activities. Blood glucose level can
data is shared with the medical experts for proper treatment fall during physical activity and exercise or quite after some
suggestions.
To examine the diabetic patients, two types of users
should be considered, aged users who remained at home
and an active user. Type II diabetes mellitus is strongly
associated with obesity and hypertension. The glycosylated
haemoglobin range greater than 7.0 is usually considered
as a positive diagnosis of diabetes. Table 1 shows the few
instances of a patient which the attributes date, time, code
and value. The code describes the blood glucose measure-
ments i.e. 62 for pre-supper, 33 for regular insulin dose, 34
for NPH insulin dose, 64 for pre-snack, 58 for pre-break-
fast, 60 for pre-lunch, etc. Figure 5 depicts the graph for Fig. 5  Pre-prandial blood glucose level measurements of a patient

13
Fog‑assisted personalized healthcare‑support system for remote patients with diabetes 3757

Table 2  Subject information

SubjectID 101
Gender Male
Age (years) 27
Height (cm) 182
Weight (kg) 83
Resting heart rate (bpm) 75
Maximum heart rate (bpm) 193
Dominant hand Right

Fig. 6  Physical activities carried out by a patient

Table 3  Physical activity monitoring

Physical activities Duration
(in seconds)

Lying 271.86
Sitting 234.79
Standing 217.16
Walking 222.52
Running 212.64
Cycling 235.74
Nordic walking 202.64
Watching TV 836.45 Fig. 7  Calories consumed per activity
Car driving 545.18
Walking upstairs 158.88
Walking downstairs 148.97
Vacuum cleaning 229.4
Ironing 235.72
Folding laundry 271.13
House cleaning 540.88
Rope jumping 129.11

time. By calculating the amount of calories consumed per

activity and subtracting it from the total amount of intake Fig. 8  Efficiency of alert generation
will help the doctor or nutritionist to recommend how many
more calories is to burn. Tables 2 and 3 presents the personal prevent disease like hypertension, diabetes, cardiovascular
information and activities performed by a subject in seconds, disease, etc. The data is then shared with the domain experts
respectively. for personalized suggestions and recommendations. In addi-
A normal person requires around 2500 kcal per day to tion, the activities performed alongside with the amount of
maintain a healthy lifestyle. Physical activities and diet are calories consumed are accessed by the authorized domain
the major sources to reduce the blood glucose level and experts continuously for appropriate suggestions and recom-
thereby increases the production of insulin by the pancreas. mendations. Figure 7 shows the activities performed and the
Depending upon the type and duration of physical activities amount of calories burned per activity in a particular inter-
performed, the amount of calories burned per intake is calcu- val of time. As a total, activities performed by a subject101
lated. A chart for 18 different physical activities performed are 4693.07 s. After analyzing the physical activities and
by a subject101 over a period described in PAMAP2_Data- amount of calories burnt, the patient is suggested to alter
set is shown in Fig. 6. Subject101 did not undergo any com- the food intake and increase the activity for some more time.
puter work, and he did not play soccer too. Let us assume The proposed fog-assisted healthcare system is examined
that the patient consumed 2500 kcal and his physical activi- systematically to verify the efficiency of alert generation.
ties burned only 1617 kcal. Hence, he needs to burn the An important parameter used to estimate the efficiency of
remaining 883 kcal to maintain a healthy lifestyle and to alert generation is the response time. Some other statistical

13

3758 M. Devarajan et al.

evaluation metrics used for this purpose are precision, sen- sensitive information, it is susceptible to privacy and secu-
sitivity, and specificity. The Higher value of precision, sen- rity issues. Thus the need for the research on robust cryp-
sitivity, and specificity shows the enhanced performance of tography techniques to preserve users’ sensitive information
the proposed system. Figure 8 shows the efficiency of gen- without compromising the performance and efficiency of the
erating timely and true emergency alerts based on response system is very high. The study will be extended to provide
time. The efficiency of alert generation is estimated based timely and user-centric recommendations by exploiting con-
on the number of delayed alerts generated to the total num- textual information and domain expert’s knowledge for the
ber of alerts generated. Delay time is computed as the time registered users.
taken to generate an alert from the occurrence of the event.
The graph shows that the latency rate for cloud computing is
more when compared to the fog computing system. In cloud 8 Conclusion
computing, a huge volume of information is transmitted for
data processing resulting in network congestion. But due to Due to the development of information and communication
the close proximity of fog computing, communication and technologies, it is feasible to monitor the blood glucose level
computation are performed at the edge of the network. It in real time to control diabetes and prevent serious conse-
reduces the network traffic and bandwidth resulting in low quences. In this work, we proposed a fog-assisted health-
latency. care system for remote patients with diabetes. The presented
Fog computing as an intermediate layer plays a vital role healthcare system connects the Internet of Things to the
in the remote patient health monitoring system. Since health cloud server through smartphones for real-time monitoring
data is processed in the intermediate fog layer, it enhances of physiological signals and contextual information continu-
the performance and efficiency of the health monitoring sys- ously. We have employed J48Graft decision tree classifier
tem. The performance gain is achieved by removing redun- to make an effective prediction on the risk level of blood
dant information and unnecessary communication. The glucose. J48Graft yields enhanced classification accuracy
execution time is reduced with respect to low latency and of 98.56% over other baseline approaches. By leveraging
network bandwidth. The latency rate is estimated in terms of fog computing as an intermediary layer, local data storage,
time taken to classify the users’ diabetes risk level. mobility, scalability, and interoperability are achieved. In
addition, it extends the cloud computing services to the edge
7.4 Medical significance of the network and facilitates the emergency alert genera-
tion. Experimental results are compared with the cloud com-
The degree of suffering from diabetes increases day by day, puting healthcare system to verify the efficiency of the fog
and it is considered as a common disease in the present day computing-based approach. The experimental results portray
scenario. One of the 15 leading causes of premature mor- the bandwidth efficiency, low latency and higher classifica-
tality rate is diabetes. People with any age and gender can tion accuracy of fog over cloud computing.
have diabetes, and it causes cardiac arrhythmia, heart attack,
kidney failure, stroke, and blindness. This is due to the vari- Acknowledgements  Authors are grateful to the SASTRA Deemed Uni-
ation of the glucose level in the blood stream. When the versity, Thanjavur, India for the financial support and infrastructural
glucose level goes down its normal range, then it is called as facilities provided to carry out this research.
hypoglycaemia, when it rises above, it is referred as hyper-
glycaemia. Though medications are available it cannot be
cured completely, the only solution is to control the level References
of glucose level in the blood stream to avoid serious conse-
quences. It necessitates continuous monitoring and regulat- Abawajy JH, Hassan MM (2017) Federated internet of things and cloud
computing pervasive patient health monitoring system. IEEE
ing the insulin level in real-time to lead a healthy lifestyle.
Commun Mag 55(1):48–53
Adeyemo OO, Adeyeye TO, Ogunbiyi D (2015) Comparative study of
7.5 Limitations and future directions ID3/C4.5 decision tree and multilayer perceptron algorithms for
the prediction of typhoid fever. Afr J Comput ICT 8(1):103–112
Ahmad M, Amin MB, Hussain S, Kang BH, Cheong T, Lee S (2016)
The fog-assisted healthcare support system is proposed to
Health Fog: a novel framework for health and wellness applica-
identify the risk level of diabetes using J48Graft classifier. tions. J Supercomput 72(10):3677–3695
However, there are several challenges to be addressed in the Ahmad A, Khan M, Paul A, Din S, Rathore MM, Jeon G, Choi GS
future. The proposed approach is designed only for diabetes (2018) Toward modeling and optimization of features selection
in big data based social Internet of Things. Future Gener Comput
and it has to be improved to make effective assessment on
Syst 82:715–726
several diseases simultaneously based on influential features Arunkumar S, Vairavasundaram S, Ravichandran KS, Ravi L
and risk factors. As the proposed approach uses patients’ (2019) RIWT and QR factorization based hybrid robust image

13
Fog‑assisted personalized healthcare‑support system for remote patients with diabetes 3759

steganography using block selection algorithm for IoT devices. Jeon G, Anisetti M, Wang L, Damiani E (2016) Locally estimated
J Intell Fuzzy Syst. https​://doi.org/10.3233/JIFS-16998​4 heterogeneity property and its fuzzy filter application for dein-
Asaithambi S, Rajappa M, Ravi L (2019) Optimization and control terlacing. Inf Sci 354:112–130
of CMOS analog integrated circuits for cyber-physical systems Kalmegh S (2015) Analysis of WEKA data mining algorithm REP-
using hybrid grey wolf optimization algorithm. J Intell Fuzzy Tree, simple CART and RandomTree for classification of Indian
Syst. https​://doi.org/10.3233/JIFS-16998​1 news. Int J Innov Sci Eng Technol 2(2):438–446
Babar M, Rahman A, Arif F, Jeon G (2018) Energy-harvesting based Kiran MS, Rajalakshmi P, Bharadwaj K, Acharyya A (2014) Adap-
on internet of things and big data analytics for smart health tive rule engine based IoT enabled remote health care data
monitoring. Sustain Comput Inf Syst 20:155–164 acquisition and smart transmission system. In: Internet of
Bonomi F, Milito R, Zhu J, Addepalli S (2012) Fog computing and Things (WF-IoT), 2014 IEEE world forum on IEEE, pp 253–258
its role in the internet of things. In: Proceedings of the first edi- Logesh R, Subramaniyaswamy V (2017) Learning recency and
tion of the MCC workshop on mobile cloud computing. ACM, inferring associations in location based social network for emo-
pp 13–16 tion induced point-of-interest recommendation. J Inf Sci Eng
Bonomi F, Milito R, Natarajan P, Zhu J (2014) Fog computing: 33(6):1629–1647
a platform for internet of things and analytics. Big data and Logesh R, Subramaniyaswamy V (2019) Exploring hybrid recom-
Internet of Things: a roadmap for smart environments. Springer mender systems for personalized travel applications. Cogni-
International Publishing, New York, pp 169–186 tive informatics and soft computing. Springer, Singapore, pp
Chehri A, Mouftah H, Jeon G (2010) A smart network architecture 535–544
for e-health applications. Intelligent interactive multimedia sys- Logesh R, Subramaniyaswamy V, Vijayakumar V, Li X (2018a) Effi-
tems and services. Springer, Berlin, pp 157–166 cient user profiling based intelligent travel recommender system
Devarajan M, Ravi L (2018) Intelligent cyber-physical system for for individual and group of users. Mob Netw Appl. https​://doi.
an efficient detection of Parkinson disease using fog com- org/10.1007/s1103​6-018-1059-2
puting. Multimed Tools Appl. https​: //doi.org/10.1007/s1104​ Logesh R, Subramaniyaswamy V, Vijayakumar V, Gao XZ, Indragan-
2-018-6898-0 dhi V (2018b) A hybrid quantum-induced swarm intelligence
Devarajan M, Fatima NS, Vairavasundaram S, Ravi L (2019) Swarm clustering for the urban trip recommendation in smart city. Future
intelligence clustering ensemble based point of interest recom- Gener Comput Syst 83:653–673
mendation for social cyber-physical systems. J Intell Fuzzy Syst. Logesh R, Subramaniyaswamy V, Malathi D, Sivaramakrishnan N,
https​://doi.org/10.3233/JIFS-16999​1 Vijayakumar V (2019) Enhancing recommendation stability of
Dziak D, Jachimczyk B, Kulesza WJ (2016) Wirelessly interfacing collaborative filtering recommender system through bio-inspired
objects and subjects of healthcare system–IoT approach. Elek- clustering ensemble method. Neural Comput Appl. https​://doi.
tronika ir Elektrotechnika 22(3):66–73 org/10.1007/s0052​1-018-3891-5
Dziak D, Jachimczyk B, Kulesza WJ (2017) IoT-based information Luo S, Ren B (2016) The monitoring and managing application of
system for healthcare application: design methodology approach. cloud computing based on Internet of Things. Comput Methods
Appl Sci 7(6):596 Prog Biomed 130:154–161
Ghanavati S, Abawajy JH, Izadi D, Alelaiwi AA (2017) Cloud-assisted Maillo J, Ramírez S, Triguero I, Herrera F (2017) kNN-IS: an iterative
IoT-based health status monitoring framework. Clust Comput spark-based design of the k-nearest neighbors classifier for big
1843:1–11 data. Knowl Based Syst 117:3–15
Gia TN, Jiang M, Rahmani AM, Westerlund T, Liljeberg P, Tenhunen Malathi D, Logesh R, Subramaniyaswamy V, Vijayakumar V, Sangaiah
H (2015) Fog computing in healthcare internet of things: a case AK (2019) Hybrid reasoning-based privacy-aware disease predic-
study on ecg feature extraction. In: Computer and information tion support system. Comput Electr Eng 73:114–127
technology; ubiquitous computing and communications; depend- Mathers CD, Loncar D (2006) Projections of global mortality and bur-
able, autonomic and secure computing; pervasive intelligence and den of disease from 2002 to 2030. PLoS Med 3(11):e442
computing (CIT/IUCC/DASC/PICOM), 2015 IEEE international Natarajan S, Vairavasundaram S, Ravi L (2019) Optimized fuzzy-based
conference on IEEE, pp 356–363 group recommendation with parallel computation. J Intell Fuzzy
Gia TN, Dhaou IB, Ali M, Rahmani AM, Westerlund T, Liljeberg P, Syst. https​://doi.org/10.3233/JIFS-16997​7
Tenhunen H (2018) Energy efficient fog-assisted IoT system for Panigrahi R, Borah S (2018) Rank allocation to J48 group of decision
monitoring diabetic patients with cardiovascular disease. Future tree classifiers using binary and multiclass intrusion detection
Gener Comput Syst 93:198 datasets. Proced Comput Sci 132:323–332
Hassan MM, Lin K, Yue X, Wan J (2017) A multimedia healthcare data Pinjari H, Paul A, Jeon G, Rho S (2018) Context-driven mobile learn-
sharing approach through cloud-based body area network. Future ing using fog computing. In: 2018 international conference on
Gener Comput Syst 66:48–58 platform technology and service (PlatCon), IEEE, pp 1–6
Hayashi Y, Tanaka Y, Takagi T, Saito T, Iiduka H, Kikuchi H, Bolo- Rathore MM, Ahmad A, Paul A, Rho S (2016a) Urban planning and
gna G, Mitra S (2016) Recursive-rule extraction algorithm with building smart cities based on the internet of things using big data
J48graft and applications to generating credit scores. J Artif Intell analytics. Comput Netw 101:63–80
Soft Comput Res 6(1):35–44 Rathore MM, Ahmad A, Paul A, Wan J, Zhang D (2016b) Real-time
Jagadeeswari V, Subramaniyaswamy V, Logesh R, Vijayakumar V medical emergency response system: exploiting IoT and big data
(2018) A study on medical Internet of Things and big data in for public health. J Med Syst 40(12):283
personalized healthcare system. Health Inf Sci Syst 6(1):14 Rathore MM, Paul A, Ahmad A, Jeon G (2017) IoT-based big data:
Jangiti S, Sri Ram E, Ravi L, Sriram VS (2019) Scalable hybrid and from smart city towards next generation super city planning. Int J
ensemble heuristics for economic virtual resource allocation in Semant Web Inf Syst (IJSWIS) 13(1):28–47
cloud and fog cyber-physical systems. J Intell Fuzzy Syst. https​:// Ravi L, Vairavasundaram S (2016) A collaborative location based travel
doi.org/10.3233/JIFS-17900​4 recommendation system through enhanced rating prediction for
Jeon G, Anisetti M, Lee J, Bellandi V, Damiani E, Jeong J (2009) the group of users. Comput Intell Neurosci 2016:1291358
Concept of linguistic variable-based fuzzy ensemble approach: Ravi L, Vairavasundaram S, Palani S, Devarajan M (2019) Location-
application to interlaced HDTV sequences. IEEE Trans Fuzzy based personalized recommender system in the internet of cultural
Syst 17(6):1245–1258 things. J Intell Fuzzy Syst. https​://doi.org/10.3233/JIFS-16997​3

13

3760 M. Devarajan et al.

Robinson RT, Harris ND, Ireland RH, Lee S, Newman C, Heller SR healthcare system. J Supercomput. https​://doi.org/10.1007/s1122​
(2003) Mechanisms of abnormal cardiac repolarization during 7-018-2331-8
insulin-induced hypoglycemia. Diabetes 52(6):1469–1474 Vairavasundaram S, Varadharajan V, Vairavasundaram I, Ravi L (2015)
Sajid A, Abbas H (2016) Data privacy in cloud-assisted healthcare sys- Data mining-based tag recommendation system: an overview.
tems: state of the art and future challenges. J Med Syst 40(6):155 Wiley Interdiscip Rev Data Min Knowl Discov 5(3):87–112
Sankar H, Subramaniyaswamy V, Vijayakumar V, Arun Kumar S, Vedanthan R, Fuster V, Fischer A (2012) Sudden cardiac death in low-
Logesh R, Umamakeswari A (2019) Intelligent sentiment analysis and middle-income countries. Glob Heart 7(4):353–360
approach using edge computing-based deep learning technique. Vijayakumar V, Malathi D, Subramaniyaswamy V, Saravanan P,
Softw Pract Exp. https​://doi.org/10.1002/spe.2687 Logesh R (2018) Fog computing-based intelligent healthcare sys-
Sarkar S, Chatterjee S, Misra S, Kudupudi R (2017) Privacy-aware tem for the detection and prevention of mosquito-borne diseases.
blind cloud framework for advanced healthcare. IEEE Commun Comput Hum Behav. https​://doi.org/10.1016/j.chb.2018.12.009
Lett 21(11):2492–2495 Wang H, Wang Z, Domingo-Ferrer J (2017) Anonymous and secure
Selvan NS, Vairavasundaram S, Ravi L (2019) Fuzzy ontology-based aggregation scheme in fog-based public cloud computing. Future
personalized recommendation for internet of medical things with Gener Comput Syst 78:712
linked open data. J Intell Fuzzy Syst. https​://doi.org/10.3233/ Wang J, Wu J, Wu Z, Anisetti M, Jeon G (2018) Bayesian method
JIFS-16996​7 application for color demosaicking. Opt Eng 57(5):053102
Sharma V, Song F, You I, Atiquzzaman M (2017) Energy efficient Wu J, Anisetti M, Wu W, Damiani E, Jeon G (2016) Bayer demosaick-
device discovery for reliable communication in 5G-based IoT ing with polynomial interpolation. IEEE Trans. Image Process
and BSNs using unmanned aerial vehicles. J Netw Comput Appl 25(11):5369–5382
97:79–95 Wu T, Wu F, Redouté JM, Yuce MR (2017) An autonomous wireless
Shi J, Wu J, Anisetti M, Damiani E, Jeon G (2017) An interval type-2 body area network implementation towards IoT connected health-
fuzzy active contour model for auroral oval segmentation. Soft care applications. IEEE Access 5:11413–11422
Comput 21(9):2325–2345 Yang G, Xie L, Mäntysalo M, Zhou X, Pang Z, Da Xu L, Zheng LR
Subramaniyaswamy V, Logesh R (2017) Adaptive KNN based recom- (2014) A health-IoT platform based on the integration of intelli-
mender system through mining of user preferences. Wirel Pers gent packaging, unobtrusive bio-sensor, and intelligent medicine
Commun 97(2):2229–2247 box. IEEE Trans Ind Inf 10(4):2180–2191
Subramaniyaswamy V, Logesh R, Abejith M, Umasankar S, Umam- Yannuzzi M, Milito R, Serral-Gracià R, Montero D, Nemirovsky M
akeswari A (2017a) Sentiment analysis of tweets for estimating (2014) Key ingredients in an IoT recipe: fog computing, cloud
criticality and security of events. J Organ End User Comput computing, and more fog computing. In: 2014 IEEE 19th inter-
(JOEUC) 29(4):51–71 national workshop on computer aided modeling and design of
Subramaniyaswamy V, Logesh R, Chandrashekhar M, Challa A, Vijay- communication links and networks (CAMAD). IEEE, pp 325–329
akumar V (2017b) A personalised movie recommendation system Yaqoob I, Ahmed E, ur Rehman MH, Ahmed AIA, Al-garadi MA,
based on collaborative filtering. Int J High Perform Comput Netw Imran M, Guizani M (2017) The rise of ransomware and emerg-
10(1–2):54–63 ing security challenges in the Internet of Things. Comput Netw
Subramaniyaswamy V, Logesh R, Indragandhi V (2018a) Intelligent 129:444–458
sports commentary recommendation system for individual cricket
players. Int J Adv Intell Paradig 10(1–2):103–117. https://doi. Publisher’s Note Springer Nature remains neutral with regard to
org/10.1504/IJAIP.2018.089492 jurisdictional claims in published maps and institutional affiliations.
Subramaniyaswamy V, Manogaran G, Logesh R, Vijayakumar V,
Chilamkurti N, Malathi D, Senthilselvan N (2018b) An ontol-
ogy-driven personalized food recommendation in IoT-based

13