Professional Documents
Culture Documents
Turabieh2019 PDF
Turabieh2019 PDF
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2926321, IEEE Internet of
Things Journal
IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. X, AUGUST 201X 1
Abstract—Data delivery and acquisition are the main factors few years. With these advancements in technologies and due
needed for the success of any proposed Internet of Medical to the extremely big number of connected devices in IoT,
Things (IoMT) systems. To achieve good performance and high transferring data between applications need to accomplished
quality of services in IoMT systems, data acquisition and delivery
should be performed accurately. In general, IoMT systems are using fast, secure and accurate mechanisms [5], [6]. Moreover,
usually vulnerable to the collected data with missing value(s) the performance of IoT applications can be enhanced by
since missing data is the main problem that affects the overall reducing the used power to transmit data between connected
performance of any system. This leads to a reduction in the devices, increase the coverage area and secure the protocol
satisfaction level of end users. Missing data for IoMT systems used for connecting more devices through 5G networks.
originates from a number of sources including bad connections,
outside attack or sensing errors. To obtain a high performance Due the large variety of the connected devices and sensors
in such systems, missing data should be imputed once occurred. in IoT applications, a large quantity of data is collected to offer
In this research paper, a Dynamic Adaptive Network-Based an acceptable and accurate service to the end users [7]–[9]. To
Fuzzy Inference System (D-ANFIS) approach is proposed to make the idea clearer, let us consider the e-Health applications
impute the missing values in a simple yet accurate manner. as an example, which have many agents (electronics, software,
The major contribution is to impute the missing value(s) once
received by dividing the collected data into two groups: complete sensors, actuators and connectivity mechanisms) that facilitate
dataset (without missing data) and incomplete dataset (with the exchange of information between the concerned people
missing data). A holdout method is used to train the D-ANFIS (e.g., doctors, patients), and maybe institutions. The use of
using complete data, while the incomplete dataset is used to such application to exchange information instantly and rapidly
impute the missing value(s). Two methods are used to evaluate helps in decreasing the cost of medical services and expand
the final performance of IoMT application: (i) ANFIS with
Genetic Algorithm (ANFIS-GA), and (ii) ANFIS with Particle the scope of medical facilities which enhances its the service
Swarm Optimization (ANFIS-PSO). The results show that the quality and enables the use of home medical care for patients.
performance of IoMT is enhanced 5% using ANFIS-GA and 3% Generally speaking, the set of the medical devices, equip-
using ANFIS-PSO. ment, and applications that can communicate online via com-
Index Terms—Missing data, ANFIS, IoMT, Particle Swarm puter networks are called Internet of Medical Things (IoMT)
Optimization, Genetic Algorithm. [10]–[12]. The online connection between devices in IoMT
has many advantages; like saving a huge amount of time
for specialists by providing instant analysis of the patients
I. I NTRODUCTION
situation, getting the results of some earlier tests like blood
With the power of the telecommunication technology; bil- sugar level test or monitoring electrocardiogram readings for
lions of things (devices) become connected via internet, which chest pain, which can be provided at-home, in a surgery or
make the life easier with the set of services they provide in a in a clinic. This helps in providing a quick response and the
wide variety of fields (i.e. environment, home and buildings, most suitable action strategy for the individual's home health
transportations, industries, etc). This phenomena is expressed care.
as Internet of Things (IoT) [1], [2]. Most of the offered Based on the earlier discussion, we noticed that the IoMT
services are in the form of applications that are built on these applications depend on the collected data from sensors and
devices. With the use of 4G networks and 5G networks in data acquisition tools, as well as on the transition medium.
the forthcoming years, IoT applications recently used either This makes those applications vulnerable to produce inac-
in commercial or personal use [3], [4]. In 5G networks, the curate results and low performance due to many reasons,
focus is on the devices rather than on the network itself including but not limited such as hardware problems ( i.e.
as in the previous versions of networks (4G or 3G). Thus, sensor problems and network devices problems ) or human
a huge number of devices will be connected in the next problems (i.e outside attackers or programmer errors). The
previously mentioned problems, and especially the sensor
Hamza is with Department of Information Technology, CIT collage, Taif
Universtiy, Taif, KSA email: h.turabieh@tu.edu.sa failure problem, cause serious problems to the IoMT. The
Majdi is with Department of Computer Science, Birzeit University, Birzeit, main serious problem for IoT systems is an absent of some
Palestine email: mmafarja@birzeit.edu important information that is required for the continuation
Seyedali is with Institute of Integrated and Intelligent Systems,
Griffith University, Nathan, Brisbane, QLD 4111, Australia email: of those applications. Missing data is a major problem that
seyedali.mirjalili@griffithuni.edu.au the causes the failure of any application and decreases its
2327-4662 (c) 2019 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/JIOT.2019.2926321, IEEE Internet of
Things Journal
IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. X, AUGUST 201X 2
performance [13], [14]. Therefore, developing a good strategy to listwise deletion (remove all missing records), pairwise
to recover the missing data is mandatory to ensure that IoMT deletion (use the correlation matrix to analysis the missing val-
systems working correctly and accurately. In Figure 1, a ues), single imputation (randomly imputation process based on
general model of a general IoMT application is illustrated. similar record), and model-based imputation (analyse samples
As an example, if a problem (i.e sensor fault) occurred in the of data to impute missing data). Haji-Maghsoudi et al. Magh-
motion sensor, then intended application will not receive the soudi2018 investigated the use of Tree-Based Models (TBMs)
sensor’s data properly, and serious problems will be caused to that based on surrogate approach (SUR) to impute missing
the end users (e.g., doctors) since they missed important data value(s) for a selected medical case study. They compared the
about the patients situation. proposed approach with the statistical methods. The authors
The necessity to manipulate the missing value(s) in IoMT enhanced the performance after imputing the missing data. In
applications to overcome the potential problems that may be addition, Suhani et al. [19] applied a hybrid approach between
occurred, motivated our attempts to propose a new approach fuzzy k-nearest neighbour and partial distance approach for
that tends to fill the missing data based on dynamic ANFIS, imputing missing value(s). In this approach, the entropy was
which considers the historical complete data that collected used to evaluate each record that contains missing values, then
by a specific sensors to predict new possible missed val- the records with entropy values less than a threshold were
ues. The proposed approach is tested on two well known removed, while other missing values (for those records with
medical datasets:( Breast cancer dataset which is known as entropy values higher than a threshold) were estimated using
Mammographic Mass dataset [15], [16], and Hepatitis dataset modied kernel-based fuzzy k-nearest neighbour algorithm. As
[17]), that simulate IoMT applications. Both datasets suffers a result, the imputation process enhance the total performance
from missing values that affect the performance of the IoMT of the medical application. However, removing some records
systems. Imputing missing data is an important issue to ensure from medical datasets may be critical for some IoMT appli-
IoMT systems keep running and provide an excellent quality cations.
of service to the customers. A latent factor approach was employed to recover (impute)
The structure of this paper is organized in seven sections: the missing data in a real medical dataset obtained from a
Section II presents the literature of missing data imputation hospital in central China in the period of 2013-2015 [20].
methods. Section III explores the datasets used in this paper. This approach was combined with a convolutional neural net-
Section IV shows the proposed approach to impute missing work based multi-modal disease risk prediction (CNN-MDRP)
data. Section V presents the performance measurements. In approach. The performance of CNN-MDRP was improved
Section VI, the experimental setup and the discussion and after recovering the missing data using latent factor model.
comparisons of the obtained results are presented. Section VII In addition, another approach was proposed in [21]; The
presents the findings and conclusion of this work. authors impute missing data for Haiti earthquake dataset and
investigates the affect of missing information on environmental
II. L ITERATURE REVIEW systems.
For any IoT systems, data is the main component of these
systems. As a result, data should be completely correct without B. Imputation of missing data for IoT
any missing value(s), and no delay while providing services As discussed earlier, having missing data in a dataset has
to the end users are the main aspects that play vital roles in a great impact on the IoT applications, and may cause many
enhancing the performance of IoMT applications. All these problems that degrade the performance and cause a serious
aspects are vulnerable to be destroyed when some data could problems. In literature, many exists literatures proposed to
be missed, mainly when the wireless devices are used for solve missing data problem for IoT systems. Fekade et al.
data collection. This is due to some reasons including but [13] applied a machine learning algorithms such as : support
not limited to intrusion attacks, connections errors, or sensors vector machine (SVM) and deep neural network (DNN) to
fault. In literature, there exist vast literature which investigate impute the missing information. The obtained results proved
the imputation methods for missing data. In the following the effectiveness of recovering the missing data in improving
subsections, we focus on the related works regarding missing the performance of the prediction methods based on two
data problem. different performance measurements (accuracy and root mean
square error). Another example for a successful approach
A. Imputation of missing data that efficiently impute the missing value(s) was investigated
Unfortunately, there is no standard approach for dealing by Fisher et al. [14]. A markov models were applied to
with the missing data. Usually, the missing values are denoted impute missing value(s) for mammographic mass dataset.
by adding some terms like N/A, missing, ?, and so on. Some Turabieh et al. [22] proposed a dynamic layered recurrent
researchers just ignore the missing data if it formulate less than neural network (D-LRNN) to impute the missing data for
5% of the whole dataset. Whilst using this approach when the IoMT applications.The authors solved two medical cases that
percentage of missing data is high, is not applicable, and a simulates real IoT application and the performance of IoMT
recovering strategy should be applied carefully [18]. application is improved after recover the missing data.
Different approaches have been discussed by different re- Chen et al. [23] solved incomplete data classification prob-
searchers to solve the missing data problem in real appli- lem by combining a kernel function with asymmetric intrin-
cations. Examples of these approaches are but not limited sic mapping. The obtained results prove that the proposed
2327-4662 (c) 2019 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/JIOT.2019.2926321, IEEE Internet of
Things Journal
IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. X, AUGUST 201X 3
approach outperforms others on two big-data case studies. and hepatitis dataset [17], [22]. Both datasets have missing
In 2017, the quality of collected data for IoT systems was value(s) that simulate real cases from medical fields.
investigated in Sat [24] for modern cities in the field of
healthcare systems. The collected data were grouped into three
categories: missing, perfect and ambiguous data. A model
was employed to handle imperfect data that upgrade the A. Mammographic mass dataset
performance of IoT systems.
Breast cancer screening usually done by mammography,
An ST-correlated proximate method was proposed in [25]
which is the most efficient methods to detect breast cancer.
to impute the collected data that has missing value (s) for
The mammographic Mass dataset is used as IoMT dataset by
IoT system. The authors applied the method on an air pol-
several researchers [14], [22]. The dataset has 961 instances
lution dataset and showed that their method provide superior
with 5 attributes for each instance and one class attribute
performance. Another work was done by Arlitt et al. [26],
(Severity) benign or malignant. The dataset has 516 benign
that investigates the effects of having missing value(s) on IoT
instances and 445 malignant instances. Table I shows the 5
systems. The percentage of missing data was reported to be
features and their expected values [16]. It has 830 (86.37%)
1%in their application.
instances with complete data and 131 instances (13.63%) with
There types of missing date were studied by Yan et al.
missing value(s).
[27] for IoT applications. To minimize their negative impacts,
the authors proposed three models including a model of
missing data imputation based on the context and linear mean TABLE I
model (MCL), binary search model (MBS), and Gaussian D ESCRIPTION OF MAMMOGRAPHIC MASS DATASET.
mixture model (MGI). All these works show the importance of
Description and expected values
considering incomplete data to minimize its negative impacts Attribute name
Data type Range
on the performance of a given system.
BI-RADS assessment Integer [1,5]
III. DATASETS Patient age Integer [18-96]
IoMT systems used to collect data that helps the medical Mass shape Integer [1,4]
staff (i.e doctors, nurses, emergency staff, etc) to take an Mass margin Integer [1,5]
accurate decisions while examining patients. The decisions of Mass density Integer [1,4]
the medical team depend on the collected data, which should
Severity Integer [0,1]
be complete and accurate. Two different datasets are used in
this paper, which are mammographic mass dataset [14], [16],
2327-4662 (c) 2019 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/JIOT.2019.2926321, IEEE Internet of
Things Journal
IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. X, AUGUST 201X 4
2327-4662 (c) 2019 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/JIOT.2019.2926321, IEEE Internet of
Things Journal
IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. X, AUGUST 201X 5
Layer 1: Every node i is represent a square node in this Layer 4: Every node in this layer is square node. Eq.(8)
layer with a node function as shown in Eq.(3) and Eq.(4) , shows the calculations of this layer, where f1 and f2 are the
where Ai and Bi are two different linguistics labels for two fuzzy if-then rules as shown in Eq.(VII) and Eq.(11), where
inputs (x and y). Usually µAi and µBi represents the mem- pi , qi , and ri are parameters set and wi is the output of layer
bership functions, which are assumed to be bell shaped with 3. All parameters referred to as consequent parameters.
a range value between 1 and 0. Eq.(5) shows the calculations
for µAi and µBi , where ai and ci are the premise parameters O4,i = w̄i · fi f or i = 1, 2 (8)
[35].
strength to the sum of all rules firing strengths as shown in 3) Dynamic ANFIS: In this paper, a new imputation ap-
Eq.(9). proach for missing data is proposed based on ANFIS. The
wi proposed approach is trained to predict the missing value(s).
O3,i = w̄i = f or i = 1, 2 (7) The proposed approach run when incomplete dataset has a
w1 + w2
2327-4662 (c) 2019 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/JIOT.2019.2926321, IEEE Internet of
Things Journal
IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. X, AUGUST 201X 6
TABLE III
ANFIS PARAMETERS SETTING .
Parameters Value
Number of layers 5
Membership function Gauss
Number of iterations 300
Momentum constant 0.9
Size of input dataset Dynamic based on number of missing data
Size of output data 1
Learning rules Least square estimations
Fig. 4. ANFIS architecture of two inputs and two rules. Gradient decent algorithm
2327-4662 (c) 2019 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/JIOT.2019.2926321, IEEE Internet of
Things Journal
IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. X, AUGUST 201X 7
TABLE IV
GA AND PSO PARAMETERS SETTING .
GA Parameters Value
Number of generations 300
Population size 50
Mutation rate 0.3
Crossover rate 0.7
Selection type Roulette wheel selection
PSO Parameters Value
Population size 50
Number of generations 300
w 1 Fig. 5. Boxplots for mammographic mass results using ANFIS-GA.
c1 2
c2 2
V. P ERFORMANCE M EASURE
Area Under the receiver operating characteristics Curve
(AUC) criteria is used to evaluate the proposed D-ANFIS
approach. A binary classification method, that predicts two
categories: faulty (0) or non-faulty (1). In General, the binary
classifier generates two possible values (errors): false negative
(FN) and false positive (FP) and two different expected values
are true negative (TN) and true positive (TP). All measure-
ments can be evaluated based on confusion matrix format for
binary classification as shown in Table V. The range of AUC
values for binary classification are explored in Table VI.
Fig. 6. Boxplots for mammographic mass results using ANFIS-PSO.
VI. R ESULTS AND DISCUSSIONS
The computer program was evaluated using MAtLAB (ver-
sion R2014a) environment by using the fuzzy toolbox. A D- B. Results of Hepatitis problem
ANFIS is used to impute the missing value(s) for two IoMT Hepatitis problem is a larger case compared to mammo-
datasets (mammographic mass and hepatitis datasets). For each graphic masses problem. Table VIII explores the obtained
problem, 11 experiments were performed. Mammographic results for using ANFIS-GA and ANFIS-PSO. It is clear that
Mass case is a simple case study due to the small number ANFIS-GA enhanced the average of AUC 5% after imputing
of attributes, while hepatitis dataset is a big case study since missing data. While ANFIS-PSO enhances the performance
it has 19 attributes. 3% based on the AUC average value. Moreover, ANFIS-GA
2327-4662 (c) 2019 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/JIOT.2019.2926321, IEEE Internet of
Things Journal
IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. X, AUGUST 201X 8
TABLE V
C ONFUSION MATRIX .
Estimated Class
Data Class = Yes Data Class = No
Actual Class Data Class = Yes True Positive (TP) False Negative (FN)
Data Class = No False Positive (FP) True Negative (TN)
TABLE VI
T HE RANGE OF AUC VALUES FOR BINARY CLASSIFIER .
Range Descriptions
AUC < 0.50 means bad results
0.50 ≤ AUC < 0.60 means poor results
0.60 ≤ AUC < 0.70 means fair results
0.70 ≤ AUC < 0.80 means acceptable results
0.80 ≤ AUC < 0.90 means excellent results
0.90 ≤ AUC ≤ 1.00 means outstanding results
R EFERENCES
Fig. 7. Boxplots for hepatitis results using ANFIS-GA.
[1] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and
The main finding of this research is that the imputation M. Ayyash, “Internet of things: A survey on enabling technologies,
protocols, and applications,” IEEE Communications Surveys Tutorials,
method integrated with prediction approaches is able to en- vol. 17, no. 4, pp. 2347–2376, Fourthquarter 2015.
hance the performance of IoMT applications. In general, [2] L. D. Xu, W. He, and S. Li, “Internet of things in industries: A survey,”
missing data decrease the performance and accuracy of IoMT IEEE Transactions on Industrial Informatics, vol. 10, no. 4, pp. 2233–
2243, Nov 2014.
systems, while imputing missing value(s) improve the perfor- [3] A. Gupta and R. K. Jha, “A survey of 5g network: Architecture and
mance of keep IoMT systems running smoothly. emerging technologies,” IEEE Access, vol. 3, pp. 1206–1232, 2015.
2327-4662 (c) 2019 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/JIOT.2019.2926321, IEEE Internet of
Things Journal
IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. X, AUGUST 201X 9
TABLE VII
S TATISTICAL MEASURES OF THE DEVELOPED MODELS FOR MAMMOGRAPHIC MASS PROBLEM .
TABLE VIII
S TATISTICAL MEASURES OF THE DEVELOPED MODELS FOR HEPATITIS PROBLEM .
[4] R. N. Mitra and D. P. Agrawal, “5g mobile technology: A survey,” of Infection and Public Health, 2018. [Online]. Available:
ICT Express, vol. 1, no. 3, pp. 132 – 137, 2015, special Issue on http://www.sciencedirect.com/science/article/pii/S1876034118301497
Next Generation (5G/6G) Mobile Communications. [Online]. Available: [18] A. Momeni, M. Pincus, and J. Libien, Imputation and Missing Data.
http://www.sciencedirect.com/science/article/pii/S2405959515300503 Springer International Publishing, 2018, pp. 185–200.
[5] M. Condoluci, G. Araniti, T. Mahmoodi, and M. Dohler, “Enabling the [19] S. Sen, M. Das, and R. Chatterjee, “Estimation of incomplete data in
iot machine age with 5g: Machine-type multicast services for innovative mixed dataset,” in Progress in Intelligent Computing Techniques: Theory,
real-time applications,” IEEE Access, vol. 4, pp. 5555–5569, 2016. Practice, and Applications, P. K. Sa, M. N. Sahoo, M. Murugappan,
[6] M. Conti, A. Dehghantanha, K. Franke, and S. Wat- Y. Wu, and B. Majhi, Eds. Singapore: Springer Singapore, 2018, pp.
son, “Internet of things security and forensics: Challenges 483–492.
and opportunities,” Future Generation Computer Systems, [20] M. Chen, Y. Hao, K. Hwang, L. Wang, and L. Wang, “Disease prediction
vol. 78, pp. 544 – 546, 2018. [Online]. Available: by machine learning over big data from healthcare communities,” IEEE
http://www.sciencedirect.com/science/article/pii/S0167739X17316667 Access, vol. 5, pp. 8869–8879, 2017.
[7] M. Haus, M. Waqas, A. Y. Ding, Y. Li, S. Tarkoma, and J. Ott, “Security [21] K. Y. Sokat, I. S. Dolinskaya, K. Smilowitz, and R. Bank,
and privacy in device-to-device (d2d) communication: A review,” IEEE “Incomplete information imputation in limited data environments with
Communications Surveys Tutorials, vol. 19, no. 2, pp. 1054–1079, application to disaster response,” European Journal of Operational
Secondquarter 2017. Research, vol. 269, no. 2, pp. 466 – 485, 2018. [Online]. Available:
[8] R. Patan, B. Rajasekhara, and K. Suresh, Performance Improvement http://www.sciencedirect.com/science/article/pii/S0377221718301310
IoT Applications Through Multimedia Analytics Using Big Data Stream [22] H. Turabieh, A. O. A. Salem, and N. Abu-El-Rub, “Dynamic l-rnn
Computing Platforms. Hershey, PA: IGI Global., 2018, pp. 200–221. recovery of missing data in iomt applications,” Future Generation Comp.
[9] P. Rizwan, M. R. Babu, B. Balamurugan, and K. Suresh, “Real-time big Syst., vol. 89, pp. 575–583, 2018.
data computing for internet of things and cyber physical system aided [23] B.-W. Chen, S. Rho, L. T. Yang, and Y. Gu, “Privacy-
medical devices for better healthcare,” in 2018 Majan International preserved big data analysis based on asymmetric imputation
Conference (MIC), March 2018, pp. 1–8. kernels and multiside similarities,” Future Generation Computer
[10] S. Rani, S. H. Ahmed, R. Talwar, J. Malhotra, and H. Song, “Iomt: A Systems, vol. 78, pp. 859 – 866, 2018. [Online]. Available:
reliable cross layer protocol for internet of multimedia things,” IEEE http://www.sciencedirect.com/science/article/pii/S0167739X16305623
Internet of Things Journal, vol. 4, no. 3, pp. 832–839, June 2017. [24] H. B. Sta, “Quality and the efficiency of data
[11] H. Silva and A. Neto, “A holistic sdn-capable session-plane tailored for in smart-cities,” Future Generation Computer Systems,
efficient iomt smart surveillance applications,” in 2016 IEEE Globecom vol. 74, pp. 409 – 416, 2017. [Online]. Available:
Workshops (GC Wkshps), Dec 2016, pp. 1–6. http://www.sciencedirect.com/science/article/pii/S0167739X16308081
[12] R. Patan, B. Rajasekhara, and K. Suresh, “Design and development of [25] I. P. S. Mary and L. Arockiam, “Imputing the missing data in iot based
low investment smart hospital using internet of things through innovative on the spatial and temporal correlation,” in 2017 IEEE International
approaches,” Biomedical Research, vol. 28, no. 11, pp. 4979–4985, Conference on Current Trends in Advanced Computing (ICCTAC),
March 2017. March 2017, pp. 1–4.
[13] B. Fekade, T. Maksymyuk, M. Kyryk, and M. Jo, “Probabilistic recovery [26] M. Arlitt, M. Marwah, G. Bellala, A. Shah, J. Healey, and B. Vandiver,
of incomplete sensed data in iot,” IEEE Internet of Things Journal, “Iotabench: An internet of things analytics benchmark,” in Proceedings
vol. PP, no. 99, pp. 1–1, 2017. of the 6th ACM/SPEC International Conference on Performance
[14] P. S. Fisher, J. James, J. Baek, and C. Kim, “Mining intelligent solution Engineering, ser. ICPE ’15. New York, NY, USA: ACM, 2015, pp. 133–
to compensate missing data context of medical iot devices,” Personal 144. [Online]. Available: http://doi.acm.org/10.1145/2668930.2688055
and Ubiquitous Computing, vol. 22, no. 1, pp. 219–224, Feb 2018. [27] X. Yan, W. Xiong, L. Hu, F. Wang, and K. Zhao, “Missing value
[Online]. Available: https://doi.org/10.1007/s00779-017-1106-1 imputation based on gaussian mixture model for the internet of things,”
[15] M. Elter, R. Schulz-Wendtland, and T. Wittenberg, “The prediction of Mathematical Problems in Engineering, vol. 2015, 2015.
breast cancer biopsy outcomes using two cad approaches that both [28] J. . R. Jang, “Self-learning fuzzy controllers based on temporal back-
emphasize an intelligible decision process.” Medical Physics, vol. 34, propagation,” IEEE Transactions on Neural Networks, vol. 3, no. 5, pp.
no. 11, pp. 4164–4172, 2007. 714–723, Sept 1992.
[16] M. Lichman. (2013) UCI machine learning repository, university of [29] E. D. Übeyli, D. Cvetkovic, G. Holland, and I. Cosic, “Adaptive neuro-
california, school of information and computer science. [Online]. fuzzy inference system employing wavelet coefficients for detection
Available: http://archive.ics.uci.edu/ml/index.php of alterations in sleep eeg activity during hypopnoea episodes,” Digit.
[17] M. Nilashi, H. Ahmadi, L. Shahmoradi, O. Ibrahim, and Signal Process., vol. 20, no. 3, pp. 678–691, May 2010. [Online].
E. Akbari, “A predictive method for hepatitis disease Available: https://doi.org/10.1016/j.dsp.2009.08.005
diagnosis using ensembles of neuro-fuzzy technique,” Journal [30] A. Sengur, “Wavelet transform and adaptive neuro-fuzzy inference
2327-4662 (c) 2019 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/JIOT.2019.2926321, IEEE Internet of
Things Journal
IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. X, AUGUST 201X 10
system for color texture classification,” Expert Systems with Seyedali Mirjalili is a lecturer at Griffith University
Applications, vol. 34, no. 3, pp. 2120 – 2128, 2008. [Online]. Available: and internationally recognized for his advances in
http://www.sciencedirect.com/science/article/pii/S0957417407000905 nature-inspired Artificial Intelligence techniques. He
[31] G. nan, A. Gktepe, K. Ramyar, and A. Sezer, “Prediction of sulfate ex- is the author of five books, 100 journal articles,
pansion of pc mortar using adaptive neuro-fuzzy methodology,” Building 20 conference papers, and 20 book chapters. With
and Environment, vol. 42, no. 3, pp. 1264–1269, 2007. over 10000 citations and H-index of 36, he is one
[32] J.-D. Wu, C.-C. Hsu, and H.-C. Chen, “An expert system of the most influential AI researchers in the world.
of price forecasting for used cars using adaptive neuro- From Google Scholar metrics, he is globally the 3rd
fuzzy inference,” Expert Systems with Applications, vol. 36, most cited researcher in Engineering Optimisation
no. 4, pp. 7809 – 7817, 2009. [Online]. Available: and Robust Optimisation using AI techniques. He
http://www.sciencedirect.com/science/article/pii/S0957417408008634 has been the keynote speaker of several international
[33] M. Liu and Y. Y. Ling, “Using fuzzy neural network approach conferences and is serving as an associate editor of top AI journals including
to estimate contractors markup,” Building and Environment, Applied Soft Computing, Applied Intelligence, IEEE Access, Advances in
vol. 38, no. 11, pp. 1303 – 1308, 2003. [Online]. Available: Engineering Software, and Applied Intelligence.
http://www.sciencedirect.com/science/article/pii/S0360132303001355
[34] M. Buragohain and C. Mahanta, “A novel approach for anfis
modelling based on full factorial design,” Applied Soft Computing,
vol. 8, no. 1, pp. 609 – 625, 2008. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S1568494607000464
[35] U. ayda, A. Hasalk, and S. Ekici, “An adaptive neuro-fuzzy inference
system (anfis) model for wire-edm,” Expert Systems with Applications,
vol. 36, no. 3, Part 2, pp. 6135 – 6139, 2009. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S0957417408004788
[36] D. E. Goldberg and J. H. Holland, “Genetic algorithms and machine
learning,” Machine Learning, vol. 3, no. 2, pp. 95–99, Oct 1988.
[Online]. Available: https://doi.org/10.1023/A:1022602019183
[37] J. H. Holland, “Genetic algorithms,” Scientific american, vol. 267, no. 1,
pp. 66–73, 1992.
[38] Y. Shi and R. C. Eberhart, “Empirical study of particle swarm optimiza-
tion,” in Evolutionary computation, 1999. CEC 99. Proceedings of the
1999 congress on, vol. 3. IEEE, 1999, pp. 1945–1950.
[39] I. C. Trelea, “The particle swarm optimization algorithm: conver-
gence analysis and parameter selection,” Information processing letters,
vol. 85, no. 6, pp. 317–325, 2003.
[40] Y. Shi and R. Eberhart, “A modified particle swarm optimizer,” in
Evolutionary Computation Proceedings, 1998. IEEE World Congress on
Computational Intelligence., The 1998 IEEE International Conference
on. IEEE, 1998, pp. 69–73.
2327-4662 (c) 2019 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.