2016 IEEE 3rd International Conference on Cyber Security and Cloud Computing
Privacy-Aware Adaptive Data Encryption Strategy
of Big Data in Cloud Computing
Keke Gai1 , Meikang Qiu2∗ , Hui Zhao3 , Jian Xiong4
Abstract—Privacy issues have become a considerable issue
while the applications of big data are growing dramatically fast
in cloud computing. The benefits us implementing these emerg-
ing technologies have improved or changed service models
and improve application performances in various perspectives.
However, the remarkably growing volume of data sizes has also
resulted in many challenges in practice. The time execution of
encrypting data is one of the serious issues during the processes
of data processing and transmissions. Many current applica-
tions abandon data encryptions in order to reach an adoptive
performance level, companions with privacy concerns. In this
paper, we concentrate on privacy issue and propose a novel
data encryption approach, named as Dynamic Data Encryption
Strategy (D2ES). Our proposed approach aims to selectively
encrypt data using privacy classification methods under timing
constraints. This approach is designed to maximize the privacy
protection scope by using a selective encryption strategy within
the required execution time requirements. The performance of
D2ES has been evaluated in our experiments, which provides
the proof of the privacy enhancement.
Index Terms—Privacy protection, data encryption, big data,
cloud computing, selective encryption strategy, time constraints
I. I NTRODUCTION
Introducing big data and cloud computing techniques has
empowered numerous applications in people’s life recently.
As an emerging technology, cloud computing has spread
into countless fields so that many new service deployments
are introduced to the public [1], [2], such as mobile parallel
computing [3]–[7] and distributed scalable data storage [8]–
[10]. Moreover, penetrations of big data techniques have
enriched the channels of gaining information from large
volume data across various platforms, domains, and systems
[11]–[13]. Being one of technical mainstreams has enabled
big data to be widely applied in multiple industrial domains
as well as explored in recent researches [14], [15].
1 K. Gai is with the Department of Computer Science, Pace University,
New York, NY 10038, USA, kg71231w@pace.edu.
2 M. Qiu is with the Department of Computer Science, Pace University,
New York, NY 10038, USA, mqiu@pace.edu.
3 H. Zhao is with the Software School, Henan University, Kaifeng,
Henan, 475000, China, zhh@henu.edu.cn.
4 J. Xiong is with the Institute of Image Communication and Net-
work Engineering (IICNE), Shanghai Jiao Tong University, China, xjar-
row@sjtu.edu.cn.
∗ M. Qiu is the corresponding author of this paper. Email address:
mqiu@pace.edu
∗∗ This work is supported by NSF CNS-1457506 and NSF CNS-
1359557. (Prof. M. Qiu).
978-1-5090-0946-6/16 $31.00 © 2016 IEEE
DOI 10.1109/CSCloud.2016.52
Authorized licensed use limited to: UNIVERSITY PUTRA MALAYSIA. Downloaded on July 12,2021 at 18:09:37 UTC from IEEE Xplore. Restrictions apply.
Despite many benefits of using big data in cloud com-
puting, there are great concerns in protecting data owners’
privacy [16], [17]. One of the privacy concerns is caused
by unencrypted data transmissions due to the large volume
of data [18], [19]. Considering an acceptable performance
level, many applications abandon using cipher texts in data
transmissions, which often occur in wireless communi-
cations. This phenomenon can result in privacy leakage
issues since plain texts are unchallenging for adversaries to
capture information in a variety of ways, such as jamming,
monitoring, and spoofing [20]. This privacy issue is exigent
because it faces to a contradiction between the security
levels and performance that is usually attached to timing
constraints.
Fig. 1: High Level Architecture of Dynamic Data Encryp-
tion Strategy (D2ES) Model
This paper addresses this issue and proposes a novel
approach that selectively encrypts data in order to maximize
the volume of encrypted data under the required timing
constraints. The proposed model is called Dynamic Data
Encryption Strategy (D2ES) model, which is designed to
protect data owners’ privacy at the highest level when
using the applicable devices and networking facilities. Fig.
1 shows the high level architecture of D2ES model, which
illustrates the main procedures and techniques. Two major
techniques used in D2ES are: (1) classifying data packages
according to privacy level and (2) determine whether data
packages can be encrypted under the timing constraints.
We design and propose an algorithm, Dynamic Encryption
Determination (DED) algorithm, which relies on the timing
constraints and facilities’ capacities to determine the data
encryption alternatives. Detailed descriptions of D2ES are
given in Section II.
Our research is significant for generating an adoptive so-
lution to protecting data owners’ privacy. Main contributions
of this work are twofold:
1) This work proposes a novel approach that selectively
273
 encrypts data packages to maximize the privacy pro- Therefore, the output is a encryption strategy that deter-
tection level under certain constraints. mines which data packages should be encrypted. Assume
e
2) The proposed algorithm offers an optimal solution that the number of encrypted data packages for Di is ND i
.
providing the maximum value of total privacy weights. The object of our research problem is maximizing the sum
3) The findings of this research provide big data-based of PWV values, which is expressed in Equation (1). Since
solutions with an adaptive approach of protecting pri- unencrypted data packages do not earn any privacy weights,
vacy. The proposed method can be also implemented only encrypted data packages are counted in our model.
in distributed storages in cloud computing. 
e
The rest of this paper is organized by the following Ouput = M ax( (ND i
× WDi )) = P (1)
order: Section II defines the main research problem and
describes operating principle of D2ES. Next, Section III The condition is the total execution time is no longer than
provides a motivational example explaining the mechanism the required timing constraint Tc . The length of Tc must
of the proposed approach in the given operating scenario. In satisfy the following requirement, as shown in Equation (2).
addition, Section IV illustrates the main algorithm designed The expression shows the minimum execution time of data
for supporting D2ES. Furthermore, our experimental con- operations, which excludes all encryptions.
figurations as well as a few experimental results are given 
n
in Section V. Finally, conclusions are drawn in Section VI. Tc  (NDi × TD i
) (2)

After implementing D2ES approach, some data packages

II. C ONCEPTS AND THE P ROPOSED A PPROACH
are selected to be encrypted. Configure that the encrypted
A. Problem Definition data set is {Dk } and the non-encrypted data set is {Dj }
We describe the main research problem in this section. The total execution time can be gained by Equation (3):
Definition II.1 shows the identified research problem that  
e e n n
Ttotal = (ND × TD )+ (ND × TD ) (3)
is Maximum Data Package under Timing Constraints (MD- k k j j

PuTC) problem. where Tc  Ttotal .

Definition II.1. Maximum Data Package Under Timing Identifying the critical problem is the fundamental of
Constraints (MDPuTC) Problem: Inputs: data package implementing D2ES model. The following section will
types {Di }, the number of data for each data package explain the main mechanism of data alternatives in our
type NDi , execution time when encrypting data for each model.
e
single data TD i
, execution time without encryptions for each
n
single data TD , the privacy weight value for each data type B. Dynamic Data Encryption Strategy (D2ES) Model
i
WDi . Outputs: a strategy determining which data will be
encrypted. The proposed problem is finding out the approach
that can gain the maximum total privacy weight value under
a given timing constraint.
As illustrated in Definition II.1, the main inputs include
five variables. First, input data include a group of packages
that are classified into different types, represented as a
set {Di }. The number of data packages in each type Di
is represented as NDi . Moreover, there are two kinds of
execution modes, which include Operation with Encryptions Fig. 2: Crucial phases of Dynamic Data Encryption Strategy
(OwE) and Operation with Non-Encryption (OwNE). The (D2ES) Model
execution time of each data package Di in OwE mode is
e
TD i
. Similarly, the execution time of each data package
n
Di in OwNE mode is TD i
. Furthermore, we introduce Based on the definitions given in Section II-A, we present
a parameter, Privacy Weight Value (PWV), for each data our D2ES model in this section. The crucial goal of D2ES
package type in order to calculate the beneficial acquisitions model is solving the problem defined in Definition II.1.
from encrypting data, represented as WDi . There are mainly three phases forming the solution. Fig.
The meaning of PWV is a criterion showing security 2 illustrates three crucial phases of D2ES model.
significance levels. The acquisitions of PWV values that 1) Phase I: Sorting by Weights: This is a preparation
categorize security issues into multiple levels can be gained phase of the model. All data package types are sorted at this
by various approaches, such as scorecard sheet [21], [22] phase. The sorting operations consider both execution time
and security measurement category [23]. In our proposed and privacy protections; thus two variables are involved,
model, the PWV value represents the privacy importance which are PWVs and the corresponding encryption execu-
for each data package. tion time. For each data package Di , the value used for

274

Authorized licensed use limited to: UNIVERSITY PUTRA MALAYSIA. Downloaded on July 12,2021 at 18:09:37 UTC from IEEE Xplore. Restrictions apply.
sorting operations is defined as SDi , which can be gained The objective is finding out the strategy that can earn
by Equation (4). the highest total PWV by choosing a set of data packages
e
for encryptions. Table I shows a mapping of data package
SDi = WDi /TD i
(4) types with the corresponding values. We name this table
as M Table. For example, D1 has 3 data packages, which
The expression shows that the efficiency of protecting
requires 5-unit time for encryptions while needs 1-unit time
privacy. The sorting operation uses a descending order. The
for non-encryption. The privacy weight value of D1 is 2.5
sorting results form a table that is named S Table.
that is higher than any other types.
2) Phase II: Data Alternatives: This phase is the crucial
step of selecting data packages for encryption operations. S TABLE I: M Table: Table mapping data types, amount, and
Table will be used for providing the reference of protection working modes. DPT (Data Package Type); M1 : Mode 1
efficiencies. The operating principle is that data package (with encryptions); M2 : Mode 2 (without encryptions)
with higher value of SDi has a higher-level alternative
priority than those data packages having lower values of M1 M2
DPT Amount e n
TD W Di TD W Di
SDi . There are a few sub-steps for selecting data packages. D1 3 5
i
2.5 1
i
0
First, a timing scope needs to be identified. The given D2 4 3 2 0.5 0
timing constraint is Tc . Therefore, the timing scope is [0, D3 2 3 1.2 0.5 0
Ts ], in which the value of Ts can be gained from Equation D4 2 4 1 1 0
(5).
First, we calculate values of SDi for the sorting purpose.
 In this case, we list results in Table II. Meanwhile, Ts = 17,
n
Ts = Tc − (NDi × TD i
) (5)
which derives from (25 - (1×3 + 0.5×4 + 0.5×2 + 1×2)).
Next, data alternatives are executed. Each encrypted data
package’s execution time is TD e
. We first encrypt the data TABLE II: S Table: Table for SDi Values
i
package with the highest SDi value. The operation will not
S D1 S D2 S D3 S D4
be ended until two situations occur. The first situation is that 0.5 0.67 0.4 0.25
all data packages are encrypted. The other situation is that
e
the execution time TD i
is longer than the rest of the time.
Define the rest of the execution time is Tr , where According to the results shown in Table II, the priority
Tr  Ts . In our model, we calculate time Tr considering sequence is SD2 →SD1 →SD3 →SD4 .
both execution time with executions and execution time
without encryptions. Once the data package is selected to TABLE III: Table for Data Alternatives Process
be encrypted, the execution time without encryption should
Tr D2 D1 D3 D4
be added to Tr . Assume that the selected data packages are
17 0 0 0 0
{Ds }. Equation (6) represents the formulas of calculating 14.5 1 0 0 0
Tr . 12 2 0 0 0
9.5 3 0 0 0
  7 4 0 0 0
e n 3 4 1 0 0
Tr = Ts − (NDs × TD s
)+ (NDs × TD s
) (6) 0.5 4 1 1 0
The data alternatives process ends when Tr is lower than
any left data package’s execution time with encryptions. Moreover, we use our DED algorithm to produce a table
3) Phase III: Output: This phase mainly output the mapping the data alternatives. Table III shows the data
results of the Phase II. An encryption plan will be generated alternatives process and the results. As shown in the table,
at this phase. In order to provide more concise presentation, we generate the following data encryption strategy: encrypt
Section III displays a motivational example. 4 D2 , encrypt 1 D1 , encrypt 1 D3 , and do not encrypt D4 .
The value of P is 11.7.
III. M OTIVATIONAL E XAMPLE A simple comparison is completed between D2ES and
We provide a motivational example of using our pro- greedy algorithm. Using greedy algorithm can generate a
posed approach in this section. The application scenario is strategy plan as follows: encrypt 3 D1 , encrypt 1 D2 , do
configured as follows: (1) There are 4 data package types, not encrypt D3 , and do not encrypt D4 . The P value is 9.5.
including D1 , D2 , D3 , and D4 . Each type has a certain Therefore, our approach’s P value is 23.2% higher than
amount of data packages and the execution time periods are greedy algorithm.
distinct. (2) Timing constraint Tc is 25. All data packages
need to be processed within 25-unit time. (3) Two working IV. A LGORITHMS
modes are included, M1 and M2 . M1 is the mode with We present the main algorithm used in our D2ES model
encryptions; M2 is the mode without encryptions. in this section, which is Dynamic Encryption Determination

275

Authorized licensed use limited to: UNIVERSITY PUTRA MALAYSIA. Downloaded on July 12,2021 at 18:09:37 UTC from IEEE Xplore. Restrictions apply.
(DED) algorithm. This algorithm is designed to dynamically TP W V is represented in Equation (7).
select data packages that can be encrypted under the certain k

condition, in which considers both timing constraints and e
TP W V = (ND i
× W Di )
facilities’ capacities. i=1
(7)
Inputs of DED algorithm include M Table, S Table, and e e e
=ND 1
W D1 + ND 2
+ ··· +
W D2 ND k
W Dk
Tc . Samples of M Table and S Table are given in Table I and
Furthermore, for understanding the complexity of MD-
II. The output is the data encryption strategy plan P that di-
PuTC problem, we propose the Maximum Privacy Weight
rects which data packages need to be encrypted. Algorithm
(MPW) Theorem. The theorem states that MDPuTC prob-
IV.1 represents the pseudo codes of DED algorithm.
lem is a NP-hard problem, which aims to calculate the max-
imum PWV total value from a group of variables, consisting
Algorithm IV.1 Dynamic Encryption Determination of encryption and non-encryption working mode’s execution
(DED) algorithm time, privacy weights, and the number of each data package
Require: ST able, M T able, Tc type. The definition of MPW theorem is shown in Theorem
Ensure: P (Encryption Strategy Plan) IV.1.
1: Input ST able, M T able, Tc Theorem IV.1. Maximum Privacy Weight Theorem: ∃ N
2: Initialize P ← ∅ types of data packages Di , each Di ’s has a privacy weight
3: Initialize endF lag ← F alse WDi when data are encrypted, the number of encrypted
4: while STable is not empty do e
data package for each Di is ND i
. TP W V is the sum of all
5: Get Di having the highest priority from STable gained privacy weights. Calculating the maximum value of
6: for ∀ Di , i=1 to NDi do TP W V is a MDPuTC problem that is NP-hard.
e n
7: if Ts > TD i
− TD i
then
8: Add one Di to P The proof of Theorem IV.1 is given below:
e n
9: Ts ← Ts − (TD i
− TD i
) Proof. ∃ N types of data packages Di , each Di ’s privacy
10: else weight WDi for the data encryptions, the number of en-
11: Break e
crypted Di is ND i
, the total privacy weights is TP W V .
12: end if For gaining the maximum value of TP W V , a Traversing
13: end for Search (TvS) with comparison is required.
14: end while
N The traversing
search can be expressed by T vS = i=1 NDi , where the
15: Output P maximum value can be gained by comparing all situations.
This proves that traversing search for MDPuTC is a NP-
The main steps of the DED algorithm are illustrated as hard.
follows: Moreover, MDPuTC problem can be reducible to Bin Pack-
ing Problem (BPP) that is a combinatorial NP-hard problem
1) Input timing constraint Tc and two tables S Table and M [24], [25]. Using the information above can construct a bin
Table. Initialize a strategy plan dataset P as an empty that refers to the total timing constraint. Each encrypted
set. Initialize a variable endFlag and assign a False data can be constructed as a package that will be filled
value to it. into the bin. The privacy weight for each data package can
2) Enter a While loop. We estimate whether the data be constructed as the value of the package. Maximizing
packages should be encrypted one by one in a sequence the total PWV value can be constructed by putting some
depending on the priority level. The data package packages from a group of packages into a fixed capacity bin
having a higher-level priority will be estimated first. in order to obtain the maximum value of package values,
3) Keep updating the execution time scope Ts . Each data such that BPPp MDPuTC. Therefore, MDPuTC problem
package’s non-encryption time needs to be added if the is also a NP-hard problem. Proved.
encryption time mode is selected during the process of
updating execution time scope. Moreover, this algorithm represents a fundamental mech-
4) Add the data package to the set P when the value of anism used in our proposed D2ES model. There will be
Ts is greater than 0. diverse implementations by adding parameters, conditions,
5) End While loop when there is no data package match- or coefficients, which are conditional on the application
ing the condition any more. environment and practical demands. The following section
6) Output the set P that consists of a set of data packages explicates our experimental evaluations as well as partial
Di . Encrypt all data packages in p. experimental findings.
In addition, the outputs of DED algorithm need to have V. E XPERIMENT AND THE R ESULTS
the maximum value of total PWVs, P. The calculation of We illustrated our experimental evaluations in this sec-
P can derivefrom Equation (1). The value of total PWVs tion. Section V-A represented our experimental configura-
e
(TP W V ) is (ND i
× WDi ). Therefore, the expression of tions. Section V-B provided partial experimental results.

276

Authorized licensed use limited to: UNIVERSITY PUTRA MALAYSIA. Downloaded on July 12,2021 at 18:09:37 UTC from IEEE Xplore. Restrictions apply.
A. Experimental Configurations
We established the experimental environment based on
our laboratorial setting. First, we developed a simulator in
order to simulate multiple approaches and evaluate their
performances. We selected Advanced Encryption Standard
(AES) as the encryption method in our experiments. Next,
cloud environments were set up in our lab as well. The
hardware included an HP server with an 8-core CPU, a
9 GM memory, with MySQL 5.7. The cloud computing
side was installed on a VMWare workstation with running
a Ubuntu 15.04 LTS Server.
Furthermore, we evaluated our D2ES model’s perfor-
mances in two perspectives. First, we measured the D2ES
model in various experimental settings, including different
data package inputs and timing constraints. Second, we also Fig. 3: Distributions of the strategy generation time between
accomplished comparisons between D2ES model with Brute D2ES and BF under Setting 1.
Force (BF) in order to examine the total privacy weight gains
and expected execution time. Detailed experimental settings
were listed as follows:
• Setting 1: We accomplished 100 simulations; the
amount of input data packages was configured between
1 and 4; the execution time length were configured
between 3 and 7 units of time; the weights of en-
cryption levels were from 1 to 6; the execution time
of non-encryption was configured from 1 to 2; non-
encryption weight was 0. Different input data packages
were applied each time.
• Setting 2: We accomplished 100 simulations; the
amount of input data packages was configured between
3 and 4; the execution time length were configured
between 8 and 10 units of time; the weights of en-
Fig. 4: Comparisons of the expected execution time and total
cryption levels were from 1 to 3; the execution time
privacy weights between D2ES and BF under Setting 1.
of non-encryption was configured from 1 to 3; non-
encryption weight was 0. Different input data packages
were applied each time.
B. Experimental Results
We illustrated a few experimental results in this section.
The displayed results were aligned with our two exper-
imental settings. Fig 3 showed the comparisons of the
strategy generation execution time between D2ES and BF
under Setting 1. The figure displayed the execution time
distributions when both D2ES and BF are generating the
strategy for the same input data package. Based on the
representation of the figure, using our approach could reduce
99.4% strategy generation time than BF. The efficiency
of generating strategy plans was fully within the accepted
scope.
Moreover, Fig. 4 showed comparisons of the expected
Fig. 5: Distributions of the strategy generation time between
execution time and total privacy weights between D2ES
D2ES and BF under Setting 2.
and BF under Setting 1. The results generated by BF were
optimal solutions. As our observation, there were slight
differences between D2ES and the optimal solutions. The
average of our approach’s expected total privacy weights Furthermore, Fig. 5 and 6 exhibited the same examina-
was 7.6% less than the optimal solutions, and the required tions under Setting 2. We had gained the similar results to
execution time was 1.98% shorter than the optimal ones. the Setting 1. According to our data collections, the average

277

Authorized licensed use limited to: UNIVERSITY PUTRA MALAYSIA. Downloaded on July 12,2021 at 18:09:37 UTC from IEEE Xplore. Restrictions apply.

Fig. 6: Comparisons of the expected execution time and total
privacy weights between D2ES and BF under Setting 2.
of D2ES’s expected total privacy weights was 0.9% less
than the optimal solutions. The expected execution time was
0.7% shorter than the optimal solutions as well.
In summary, our proposed D2ES could generate the
encryption strategy within an adaptive timing scope. The
expected privacy protection was very close to the optimal
results. The experimental results reached our design goal
and matched the theoretical results. Future research will ad-
dress the practical measurements in real-world evaluations.
VI. C ONCLUSIONS
This paper focused on the privacy issues of big data
and considered the practical implementations in cloud com-
puting. The proposed approach, D2ES, was designed to
maximize the efficiency of privacy protections. Main al-
gorithm supporting D2ES model was DED algorithm that
was developed to dynamically alternative data packages for
encryptions under different timing constraints. The experi-
mental evaluations showed the proposed approach had an
adaptive and superior performance.
R EFERENCES
[1] M. Qiu, M. Zhong, J. Li, K. Gai, and Z. Zong. Phase-change
memory optimization for green cloud with genetic algorithm. IEEE
Transactions on Computers, 64(12):3528 – 3540, 2015.
[2] K. Gai and S. Li. Towards cloud computing: a literature review on
cloud computing and its development trends. In 2012 Fourth Int’l
Conf. on Multimedia Information Networking and Security, pages
142–146, Nanjing, China, 2012.
[3] J. Li, Z. Ming, M. Qiu, G. Quan, X. Qin, and T. Chen. Resource
allocation robustness in multi-core embedded systems with inaccurate
information. Journal of Systems Architecture, 57(9):840–849, 2011.
[4] L. Chen, Y. Duan, M. Qiu, J. Xiong, and K. Gai. Adaptive resource
allocation optimization in heterogeneous mobile cloud systems. In
The 2nd IEEE International Conference on Cyber Security and Cloud
Computing, pages 19–24, New York, USA, 2015. IEEE.
[5] J. Niu, Y. Gao, M. Qiu, and Z. Ming. Selecting proper wireless
network interfaces for user experience enhancement with guaran-
teed probability. Journal of Parallel and Distributed Computing,
72(12):1565–1575, 2012.
[6] K. Gai, Z. Du, M. Qiu, and H. Zhao. Efficiency-aware workload
optimizations of heterogenous cloud computing for capacity planning
in financial industry. In The 2nd IEEE International Conference on
Cyber Security and Cloud Computing, pages 1–6, New York, USA,
2015. IEEE.
278
Authorized licensed use limited to: UNIVERSITY PUTRA MALAYSIA. Downloaded on July 12,2021 at 18:09:37 UTC from IEEE Xplore. Restrictions apply.
[7] K. Gai, M. Qiu, H. Zhao, L. Tao, and Z. Zong. Dynamic energy-aware
cloudlet-based mobile cloud computing model for green computing.
Journal of Network and Computer Applications, 59:46–54, 2015.
[8] K. Gai, M. Qiu, and H. Zhao. Security-aware efficient mass
distributed storage approach for cloud systems in big data. In The
2nd IEEE International Conference on Big Data Security on Cloud,
pages 140–145, New York, USA, 2016.
[9] K. Gai, M. Qiu, L. Tao, and Y. Zhu. Intrusion detection techniques
for mobile cloud computing in heterogeneous 5G. Security and
Communication Networks, pages 1–10, 2015.
[10] G. Wu, H. Zhang, M. Qiu, Z. Ming, J. Li, and X. Qin. A
decentralized approach for mining event correlations in distributed
system monitoring. Journal of parallel and Distributed Computing,
73(3):330–340, 2013.
[11] H. Yin and K. Gai. An empirical study on preprocessing high-
dimensional class-imbalanced data for classification. In The IEEE
International Symposium on Big Data Security on Cloud, pages
1314–1319, New York, USA, 2015.
[12] H. Liang and K. Gai. Internet-based anti-counterfeiting pattern with
using big data in china. In The IEEE International Symposium on Big
Data Security on Cloud, pages 1387–1392, New York, USA, 2015.
IEEE.
[13] H. Jean-Baptiste, M. Qiu, K. Gai, and L. Tao. Meta meta-analytics
for risk forecast using big data meta-regression in financial industry.
In The 2nd IEEE International Conference on Cyber Security and
Cloud Computing, pages 272–277, New York, USA, 2015. IEEE.
[14] Y. Li, K. Gai, Z. Ming, H. Zhao, and M. Qiu. Intercrossed
access control for secure financial services on multimedia big data
in cloud systems. ACM Transactions on Multimedia Computing
Communications and Applications, PP(99):1, 2016.
[15] H. Yin, K. Gai, and Z. Wang. A classification algorithm based on
ensemble feature selections for imbalanced-class dataset. In The
2nd IEEE International Conference on High Performance and Smart
Computing, pages 245–249, New York, USA, 2016.
[16] Y. Li, W. Dai, Z. Ming, and M. Qiu. Privacy protection for preventing
data over-collection in smart city. IEEE Transactions on Computers,
PP:1, 2015.
[17] K. Gai, M. Qiu, L. Chen, and M. Liu. Electronic health record
error prevention approach using ontology in big data. In 17th
IEEE International Conference on High Performance Computing and
Communications, pages 752–757, New York, USA, 2015.
[18] M. Qiu, K. Gai, B. Thuraisingham, L. Tao, and H. Zhao. Proactive
user-centric secure data scheme using attribute-based semantic access
controls for mobile clouds in financial industry. Future Generation
Computer Systems, PP:1, 2016.
[19] K. Gai, M. Qiu, B. Thuraisingham, and L. Tao. Proactive attribute-
based secure data schema for mobile cloud in financial industry. In
The IEEE International Symposium on Big Data Security on Cloud;
17th IEEE International Conference on High Performance Computing
and Communications, pages 1332–1337, New York, USA, 2015.
[20] L. Ma, L. Tao, Y. Zhong, and K. Gai. RuleSN: Research and appli-
cation of social network access control model. In IEEE International
Conference on Intelligent Data and Security, pages 418–423, New
York, USA, 2016.
[21] M. Baran, P. Carpenter, L. Borbye, D. Lubkeman, M. Ligett, and
D. Covington. A new professional science master program for electric
power systems engineering. IEEE Transactions on Power Systems,
29(4):1903–1910, 2014.
[22] K. Gai and A. Steenkamp. A feasibility study of Platform-as-a-
Service using cloud computing for a global service organization.
Journal of Information System Applied Research, 7:28–42, 2014.
[23] M. Qiu, W. Gao, M. Chen, J. Niu, and L. Zhang. Energy efficient
security algorithm for power grid wide area monitoring system. IEEE
Transactions on Smart Grid, 2(4):715–723, 2011.
[24] C. Cicconetti, L. Lenzini, A. Lodi, S. Martello, E. Mingozzi, and
M. Monaci. Efficient two-dimensional data allocation in IEEE 802.16
OFDMA. IEEE/ACM Transactions on Networking, 22(5):1645–1658,
2014.
[25] M. Vidyasagar. A metric between probability distributions on finite
sets of different cardinalities and applications to order reduction.
IEEE Transactions on Automatic Control, 57(10):2464–2477, 2012.