Content uploaded by Lokesh Chouhan
Author content
All content in this area was uploaded by Lokesh Chouhan on Apr 15, 2019
Content may be subject to copyright.
A Survey on Applications of Machine Learning in
Wireless Sensor Networks
Nancy Chauhan1, Vidushi Agarwal2, Amitosh Swain Mahapatra3, and Lokesh
Chouhan4
1 Indian Institute of Information Technology Una, India
nancychn1@gmail.com
2 National Institute of Technology Hamirpur, India
cs14mi506@nith.ac.in
3College of Engineering and Technology, Bhubaneswar, India
amitosh.swain@gmail.com
4National Institute of Technology Hamirpur, India
lokeshchouhan@gmail.com
Abstract. Wireless sensor networks (WSNs) are utilized to observe, explore
and control the physical world. Wireless Sensor Networks are generally de-
ployed in environments which are dynamic in nature. To avoid unnecessary re-
design, sensor networks utilize machine learning techniques. Machine learning
is also used to maximize the security, efficiency, lifetime, and resource utiliza-
tion in such networks. In this paper, we present a literature survey of various
machine learning applications that are used or are in research to address the op-
erational and non-operational challenges in WSN.
Keywords: Clustering, Data aggregation, Machine learning, operational and
non-operational challenges, Security, Wireless Sensor Networks.
1 INTRODUCTION
Wireless sensor networks are a constellation of miniature, low cost, low power and
typically autonomous sensors, connected through a common network configuration.
These nodes continuously gather data and transmit it to central control units for stor-
age or further analysis. These sensor nodes could be of any flavor, such as thermal,
acoustic, chemical, or electrical sensors. This diversity in sensors gives rise to enor-
mous possibilities for building powerful and robust applications. However, designing
such applications is a challenging task. Engineers have to tackle issues related to lo-
calization, node clustering, routing, event detection, fault diagnosis, ensuring data
integrity, and security of deployed networks.
Machine learning (ML) is defined as the techniques and algorithms used for auto-
matic acquisition of knowledge from example data or past experience to optimize a
performance criterion [1]. In the context of WSNs, we can observe that the power of
machine learning is the ability to utilize historical data to improve the efficiency of
WSNs. This is evident from the following reasons:
2
1. WSNs are deployed in unstable and dangerous environments such as high-
temperature furnaces, radiation prone areas, areas in the vicinity of corrosive
chemicals, where the odds of errors due to mechanical and electrical interference
are very high. ML techniques can help in identifying and correcting such errors.
2. Problems such as routing, while simple to define, do not have a simple mathemati-
cal model. ML can be used to design an optimum solution based on practical ob-
servations.
3. Some sensor networks capture an enormous amount of data which becomes infea-
sible to derive correlations. ML techniques have been successfully used to reduce
the dimensionality of such observations and extract useful information.
Since its introduction in the 1950s, Machine learning has been applied to a variety
of tasks and much progress has been made in the field with new inventions of algo-
rithms. Yet, applications of ML techniques to all tasks in WSNs are still a challenge
due to the limitation of computing power and memory, which pose a unique challenge
in themselves.
Traditionally, engineers working on Wireless Sensor Networks have viewed ma-
chine learning as a tool for creating prediction models. However, it is imperative for
us to recognize machine learning as a field with much greater scope. When applied to
various different problems in the field of WSNs, machine learning provides para-
mount flexibilities and improvements.
The rest of this paper is organized as follows:
In section II, we present an introduction to ML algorithms discussed in subsequent
sections.
In section III, we discuss ML applications to operational challenges in WSNs and
in section IV, we discuss non-functional challenges.
In Section V, we discuss some interesting problems which use WSNs and machine
learning but do not fall under either of the aforesaid categories.
Section VI highlights some emerging areas where ML applications are promising,
but practical implementations are at their infancy.
2 OVERVIEW OF MACHINE LEARNING ALGORITHMS
We categorize machine learning algorithms by their method of training – supervised,
unsupervised and reinforcement learning [2, pp. 5]. In supervised learning, the pro-
gram is provided with a labeled dataset and the algorithm tries to model the relation-
ship between the known predictors and the labeled criterions [2, pp.6]. Unsupervised
learning models do not have any predetermined labels provided for training. The goal
of such models is to classify the input data into different clusters based on some simi-
larity relationship [2, pp. 281–282]. Reinforcement learning models used a reward
based learning scheme. The agent is made to interact with the environment directly
and learns the set of actions that maximize long-term rewards [2, pp. 231–232].
In this section, we discuss some theoretical concepts and try to break-down some
challenges in WSNs which can be formulated as a problem to be tackled by machine
learning.
3
2.1 K-Nearest Neighbor
This algorithm classifies predictor inputs based on closeness with predictor values
provided during training examples. A simple implementation may use the Euclidean
distance between different points and classify according to the majority among the
nearest n values [2, pp. 158-160], computing which is inexpensive. Correlation of
sensor readings in a locality along with the low computational power requirements
makes it a suitable choice in WSNs. However, in [3] we see that the accuracy of KNN
degrades when the predictors are from high dimensional spaces.
2.2 Support Vector Machines
Support Vector Machine is an algorithm to construct a high-dimensional hyper-plane
from input vectors which can be then used for regression, classification or outlier
detection [4]. To illustrate the use of SVMs, we use an example of outlier detection.
We can model the input data on SVM hyper-plane and investigate the spatial and
temporal correlation. By plotting the input data on an n-dimensions and separated by
an-1 dimensional hyper-plane constructed by the SVM, we determine the outliers as
values lying on the side with lesser frequency of observations.
2.3 Neural Networks
Neural networks are inspired by the working of the human brain. We can visualize
neural networks as cascaded decision units (such as perceptions or radial basis func-
tions) with the output of one layer forming the input of the next [5]. It has been shown
that neural networks can be generalized into any nonlinear function, given enough
time and training examples [6]. However, the neural networks rely on heavy matrix
operations which are computationally very expensive. This makes it practical only to
implement centralized solutions.
To illustrate the use of neural networks, we demonstrate a simple sensor node lo-
calization problem. Node localization involves complex relationships from various
network metrics including but not limited to the angle of arrival (AOA), received
signal strength indicator (RSSI), time difference of arrival (TDOA), and time of arri-
val (TOA). Such a relationship can be learned by a neural network using a training
dataset of multiple observations.
2.4 Decision Trees
The decision tree algorithm is used for classification of input vectors through the use
of a learning tree [1, pp. 185–187]. During the learning phase, predictor variables are
compared to generate boundary conditions for each of the classes. Decision trees have
been extensively used in WSNs – such as to network reliability metrics.
2.5 Bayesian Methods
These are the family of algorithms relying on probability. Bayesian inference uses a
small training set to infer relationships a model without overfitting [7]. Bayesian
methods work with the principle of Bayes theorem in probability. An example could
be predicting the occurrence of a certain event at a time step in the future, as a result
of several past events.
4
3 OPERATIONAL CHALLENGES
There are various constraints when it comes to WSN including memory and ener-
gy, changes to topology, sudden failures and harsh environment which can be ad-
dressed using various ML techniques.
3.1 Routing issues in WSN
Limitations such as low bandwidth, limited energy and compact memory, pose major
challenges for designing a routing protocol for WSNs [8]. The routing problem can be
formulated as a graph model and the objective function can be projected as finding a
minimum cost path between the source and destination vertices [9]. The solution to
this problem is a minimum spanning tree [9], and finding such with optimal data ag-
gregation is known to be NP-Hard [10]. An ML-based approach uses prior experience
to plan optimal routing providing a solution of high accuracy.
Barbancho et al. utilized Self Organizing Maps in [11] as SIR algorithm (Sensor
Intelligence Routing). It uses SOM unsupervised learning for detecting optimized
routing paths. It is a slight moderation of the Dijkstra’s algorithm for the formation of
the backbone of the network and establishing the shortest distance paths from the base
station to all other nodes.
Guestrin et al., show the application of “Distributed Regression” in [12]. Based on
kernel linear regression, WSN nodes collaborate and optimally model a function to
individual local readings. After proper fitting, each node can answer queries for its
region or can efficiently transmit the data to a central controller.
Sun et al, show the use of “Q-Learning”, a reinforcement learning technique in de-
signing “Q-MAP” [13], to design a multicast protocol. It provides a reliable allocation
of resources. This is improved in “Q-Probabilistic Routing” by Arroyo-Valles et al
[14].
3.2 Event Recognition and Query Processing
WSN monitoring can be either continuous, or event/query driven [8]. With machine
learning, it becomes possible to assess the validity of event and restrict the query are-
as which enhance the efficiency of event detection mechanisms using limited compu-
ting and storage resources and prevents flooding the whole network.
Yu et al. [15], demonstrate forest fire detection based on neural networks. It was
shown to be cost-effective and having much better results than using satellite-based
techniques.
B. Krishnamachari and S. Iyengar in [16] show the use of decentralized Bayesian
algorithms with WSNs for identifying the distributed environmental events. Readings
beyond a certain threshold were considered as faulty.
Winter et al. [17] discuss K-NN Boundary Tree, a method for processing queries
using the k-nn algorithm. It describes the estimation of the maximum hop distance a
query can traverse from the origin point to the set of k-nearest neighbours, using re-
stricted flooding and timers to achieve efficiency. An extension, “3D KNN”, has been
proposed by Jayaraman et al. in [18] which extend this concept to a 3-dimensional
WSN.
5
Xialoi et al. [19] demonstrate the application of long short term memory, a type of
recurrent neural network, for traffic speed prediction from remote microwave sensor
data.
3.3 Data aggregation and clustering issues
Aggregating data locally and subsequently transmitting it to a sink is more energy
efficient than nodes directly transmitting data. Several cluster head (local aggregator)
selection methods have been discussed such as in [3]. Methods like similarity extrac-
tions, dimensionality reduction, and cluster election can employ ML methods for
improving clustering and aggregation which results in energy efficient WSN deploy-
ments.
Ahmed et al., show the application of decision trees in [20] in solving the problem
of cluster head election. The simulation results showed that this algorithm increases
the cluster head selection performance as compared to the LEACH algorithm.
Gaussian process (GP) is a made up of stochastic variables that are parameterized
using covariance and mean functions. In [21], Ertin described a scheme for loading
probabilistic methods of the readings based on GP regression.
The concept of distributed tracking and single target detection using sensor net-
works is addressed in [22]. “Collaborative Signal Processing” is method gathering
information from the environment under inspection. This algorithm can also be used
in tracking of multiple targets by using specific classification techniques.
Hongmei et al., [23], show the processing of self-managed clusters using neural
networks.
3.4 Object Targeting and localization
Localization is the determination of geo-coordinates of nodes. GPS (Global Position-
ing System) is a popular method for geo-location, but it is infeasible financially in
WSN nodes and is only available outdoors. Absolute locations can be predicted by
using relative locations and various techniques have been presented to improve the
performance of proximity based localization [24].
Machine learning algorithms can be used in the localization process of WSN
nodes. Few anchor points can be used to convert relative to absolute location without
the need of range measurement hardware. Machine learning can be used in surveil-
lance systems by dividing the sites to be monitored into clusters.
Li et al. [25] proposed a localization method for WSNs based on reinforcement
learning, “Dynamic Path determination of Mobile Beacons”, which is usable for real-
time analysis of mobile beacons.
Tran and Nguyen [26], proposed “LSVM”, SVM based localization algorithm
based on connection metrics such as hop count.
4 NON OPERATIONAL CHALLENGES
Apart from the functional challenges, machine learning also provides promising re-
sults for non-operational challenges such as node security and outlier detection. In this
section, we highlight some of those applications.
6
4.1 Security in Wireless Sensor Networks
Due to resource constraints, adopting strong security measures has been a long-
standing challenge for WSNs. ML techniques mitigates some attacks by detecting and
filtering misleading information. This increases the WSN lifetime by restricting the
coming of outliers and false data and maximizes network reliability.
Janakiram et al. [27] show the use of Bayesian belief networks for outlier detec-
tion. Its principle of working is to discover spatial and temporal correlations between
sensor readings in a neighborhood and infer these relationships to detect potential
outliers. This can also be used to extrapolate missing values.
Branch et al. [28] show the use of k-nearest neighbors for outlier detection by av-
eraging the reading of k-nearest nodes. However, such an approach is memory inten-
sive.
Kaplantiz et al. [29] present a packet dropping prevention using one-class SVM. It
is particularly targeted towards blackhole attacks, a kind of denial of service attack.
This technique uses bandwidth, routing, and hop count to detect malicious nodes.
Rajasegarar et al. [30] introduce a one-class quarter-sphere SVM for anomaly de-
tection. This method is computationally less intensive than traditional SVM tech-
niques. Yanh et al. [31] extended the quarter sphere SVM for online outlier detection.
SVM models perform better on the problems, but their high compute and memory
requirements limit their scalability.
Avram et al. [32] use self-organizing maps to design a method of intrusion detec-
tion in wireless ad hoc networks. Weights are learned from network metrics as input
vectors. It shows reliable performance but analyzing large-scale data is difficult due to
memory requirements.
The following Table 1 gives a summary of all the Machine Learning techniques
which can be used to overcome various security issues in WSNs.
Table 1. VARIOUS
ML T
ECHNIQUES FOR COMBATING
S
ECURITY ISSUES IN
WSN
Study
Algorithm
Complexity
Notes
Janakiram et al.[27]
Bayesian learn-
ing
Low
Extrapolates missing data
Branch et al.[28]
KNN
High
Memory intensive. Can extrapo-
late missing data.
Kaplatiz et al.[29]
SVM
High
Detects blackhole attacks
Rajasegarar et
al.[30]
SVM
High
Less communication overhead.
Avram et al. [32]
SOM
High
Reliable, but memory intensive
4.2 QOS, Fault Detection, and Data Integrity
Quality of service is used for differential priority of transmission to guarantee real-
time constraints of the sensor networks. WSNs contain multi-hop and heterogeneous
transmissions, limited by bandwidth and timing constraints. The transmitted data also
7
carries a risk of being faulty or can contain outliers. ML algorithms are used to identi-
fy data streams and simplify routing techniques. They are also employed to guarantee
data integrity and fault detection to preserve the effectiveness of WSNs with much
efficient resource utilization.
Snow et al. in [33] utilize neural networks to estimate WSN reliability. It uses sev-
eral network attributes and readings to estimate the mean time to repair (MTTR) and
mean time to failure.
In Wang et al. [34] present “Metric Map”, a framework for link quality estimation
utilizing supervised learning. It employs decision trees over several features such as
the transmission buffer size, received signal strength indicator (RSSI), and channel
load.
Ouferhat and Mellouk [35] introduce a QoS task scheduler using Q-learning tech-
nique for improving network throughput in multimedia sensor networks.
5 MISCELLANEOUS APPLICATIONS
There are certain unique application-specific challenges that cannot be classified into
popular ML WSN literature.
5.1 Reinforcement learning based resource management
“Distributed Independent Reinforcement Learning” (DIRL) [24] algorithm optimizes
various tasks by using application constraints and local information while minimizing
energy utilization. It assigns priority to different tasks by using Q-learning algorithm.
5.2 Intelligent lighting control based on neural networks
A new standard for controlling lights in smart buildings is introduced in [36] using
neural networks. A mathematical expression called “Illuminance Matrix”, is extracted
using a radial basis function network for measuring the degree of illuminance. This
scheme can produce 60% more accurate results than the standard methods.
5.3 Animals behavior classification using decision trees
WSN has many applications including habitat [37] and environmental monitoring. E.
Nadimi [38] used decision tree method to classify the active or inactive behavior of
animals by using attributes such as movement velocity or pitch angle of the neck.
5.4 Self-organizing map for clock synchronization
For operations between nodes to be consistent with each other, clock synchronization
is necessary. In [39], a method for reliable clock synchronization has been proposed to
sing self-organizing maps. Without any central timing device, nodes can predict cur-
rent time estimation using limited resources. But it assumes node deployment to be
uniform as well as equal power of transmission for all nodes.
5.5 Neural networks based air quality monitoring
Octavian Postolache [40] proposed a method to measure air pollution levels based on
neural networks using inexpensive gas nodes in which sensor readings are not affected
8
by temperature or humidity. This method uses JavaScript to distribute processing
work between end-user computers and web server.
6 EMERGING AREAS
Machine learning has been successfully applied to find solutions for a lot of problem
domains in WSNs. Yet, there are many new areas which are open and have ongoing
promising research effort. They are yet to be realized as a feasible solution. In this
section, we present an overview of several of these areas.
6.1 Distributed machine learning
Most machine learning based implementations are resource intensive for limited re-
source devices like WSNs nodes to be performed on-device. Distributed machine
learning techniques best describe these scenarios. Compared to traditional algorithms,
these algorithms utilize parallel computing and thus benefit from the individual com-
puter contributions of a WSNs swarm. Some of the recent developments include
“Adaptive Regularization of Weights” (AROW) by Crammer et al.[41], “Improved
Ellipsoid Method for Online Learning” (IELLIP) [42] and “Soft Confidence-
Weighted” (SCW).[43]
Another distributed technique is hierarchical clustering [44], an unsupervised
learning technique which can be utilized for node clustering. Clustering can be per-
formed in small groups in parallel and later merged. This information can be used for
selective activation of sensors in a cluster for purposes such as power saving, Exem-
plary methods are ”Balanced Iterative Reducing and Clustering using Hierarchies”
(BIRCH) [45] and ”Clustering Using Representatives” (CURE) [46].
6.2 Intelligent data compression and compressive sensing
It has been shown [47] that 80% of the power consumed by a WSN node is spent on
data transmission itself. This is proportional to the amount of data transmitted by each
node. Data compression of sensory or transit data will lead to reduce the amount of
transmitted data and hence reduce the power consumption.
Existing compression techniques are a tradeoff between compression ratios and
compute resources [47] both being expensive for a WSN node. Newer machine learn-
ing based compression algorithms [48] [49] provide excellent compression, but have a
very high compute requirement. Similarly, compressive sensing provides dimension
reduction, yet resource demands restrict the usefulness for on-node compression.
Methods currently being researched are component analysis, singular value decompo-
sition, and deep learning based compressive sensing etc. [50].
7 CONCLUSION
Wireless Sensor Networks, as evident from the discussion so far has a unique set of
challenges and limitations. Several of these have been shown to be solved effectively
by the use of machine learning techniques. In this survey, we have taken into consid-
eration various ML techniques to address functional, non-functional as well as appli-
cation specific challenges in WSNs. Adopting those techniques, in a real environ-
9
ment, requires careful consideration regarding the limited resources in WSNs. Future
research can be focused on hybrid machine learning techniques to mitigate the resid-
ing issues and improve the efficiency of WSNs even further. Online learning, distrib-
uted message passing, and resource management remain as open challenges.
REFERENCES
1. E. Alpaydin, Introduction to machine learning. The MIT Press, 2014.
2. S. Marsland, Machine learning: an algorithmic perspective. CRC Press, 2015.
3. K. Beyer, J. Goldstein, R. Ramakrishnan, and U. Shaft, “When is nearest neighbor mean-
ingful?” Lecture Notes in Computer Science Database Theory ICDT99, pp. 217–235,
1999.
4. C. Cortes and V. Vapnik, “Support-vector networks,” Machine Learning, vol. 20, no. 3,
pp. 273–297, 1995.
5. S. Haykin, Neural networks: a comprehensive foundation. Macmillan, 1998.
6. K. Hornik, M. Stinchcombe, and H. White, “Multilayer feedforward networks are univer-
sal approximators,” Neural Networks, vol. 2, no. 5, pp. 359–366, 1989.
7. D. Fraser, “Bayesian inference: an approach to statistical inference,” Wiley Interdiscipli-
nary Reviews: Computational Statistics, vol. 2, no. 4, pp. 487–496, 2010.
8. J. Al-Karaki and A. Kamal, “Routing techniques in wireless sensor networks: a survey,”
IEEE Wireless Communications, vol. 11, no. 6, p. 628, 2004.
9. S. K. Singh, M. Singh, D. K. Singh et al., “Routing protocols in wireless sensor networks–
a survey,” International Journal of Computer Science & Engineering Survey (IJCSES),
vol. 1, no. 2, pp. 63–83, 2010.
10. P. Eades and S. Whitesides, “The realization problem for euclidean minimum spanning
trees is np-hard,” Algorithmica, vol. 16, no. 1, pp. 60–82, 1996.
11. J. Barbancho, C. Leon, F. J. Molina, and A. Barbancho, “A new qos´ routing algorithm
based on self-organizing maps for wireless sensor networks,” Telecommunication Sys-
tems, vol. 36, no. 1-3, pp. 73–83, 2007.
12. C. Guestrin, P. Bodik, R. Thibaux, M. Paskin, and S. Madden, “Distributed regression: an
efficient framework for modeling sensor network data,” in Proceedings of the 3rd interna-
tional symposium on Information processing in sensor networks. ACM, 2004, pp. 1–10.
13. R. Sun, S. Tatsumi, and G. Zhao, “Q-map: A novel multicast routing method in wireless ad
hoc networks with multiagent reinforcement learning,” in TENCON’02. Proceedings.
2002 IEEE Region 10 Conference on Computers, Communications, Control and Power
Engineering, vol. 1. IEEE, 2002, pp. 667–670.
14. R. Arroyo-Valles, R. Alaiz-Rodriguez, A. Guerrero-Curieses, and J. CidSueiro, “Q-
probabilistic routing in wireless sensor networks,” in Intelligent Sensors, Sensor Networks
and Information, 2007. ISSNIP 2007. 3rd International Conference on. IEEE, 2007, pp. 1–
6.
15. L. Yu, N. Wang, and X. Meng, “Real-time forest fire detection with wireless sensor net-
works,” in Wireless Communications, Networking and Mobile Computing, 2005. Proceed-
ings. 2005 International Conference on, vol. 2. IEEE, 2005, pp. 1214–1217.
16. B. Krishnamachari and S. Iyengar, “Distributed bayesian algorithms for fault-tolerant
event region detection in wireless sensor networks,” IEEE Transactions on Computers,
vol. 53, no. 3, pp. 241–250, 2004.
17. J. Winter, Y. Xu, and W.-C. Lee, “Energy efficient processing of k nearest neighbor qu e-
ries in location-aware sensor networks,” in Mobile and Ubiquitous Systems: Networking
10
and Services, 2005. MobiQuitous 2005. The Second Annual International Conference on.
IEEE, 2005, pp. 281–292.
18. P. P. Jayaraman, A. Zaslavsky, and J. Delsing, “Intelligent processing of k-nearest neigh-
bors queries using mobile data collectors in a location aware 3d wireless sensor network,”
in International Conference on Industrial, Engineering and Other Applications of Applied
Intelligent Systems. Springer, 2010, pp. 260–270.
19. X. Ma, Z. Tao, Y. Wang, H. Yu, and Y. Wang, “Long short-term memory neural network
for traffic speed prediction using remote microwave sensor data,” Transportation Research
Part C: Emerging Technologies, vol. 54, pp. 187–197, 2015.
20. G. Ahmed, N. M. Khan, Z. Khalid, and R. Ramer, “Cluster head selection using decision
trees for wireless sensor networks,” in Intelligent Sensors, Sensor Networks and Infor-
mation Processing, 2008. ISSNIP 2008. International Conference on. IEEE, 2008, pp.
173–178.
21. E. Ertin, “Gaussian process models for censored sensor readings,” in Statistical Signal
Processing, 2007. SSP’07. IEEE/SP 14th Workshop on. IEEE, 2007, pp. 665–669.
22. D. Li, K. D. Wong, Y. H. Hu, and A. M. Sayeed, “Detection, classification, and tracking of
targets,” IEEE signal processing magazine, vol. 19, no. 2, pp. 17–29, 2002.
23. H. He, Z. Zhu, and E. Makinen, “A neural network model to minimize the connected dom-
inating set for self-configuration of wireless sensor networks,” IEEE Transactions on Neu-
ral Networks, vol. 20, no. 6, pp. 973–982, 2009.
24. A. K. Paul and T. Sato, “Localization in wireless sensor networks: A survey on algorithms,
measurement techniques, applications and challenges,” Journal of Sensor and Actuator
Networks, vol. 6, no. 4, p. 24, 2017.
25. S. Li, X. Kong, and D. Lowe, “Dynamic path determination of mobile beacons employing
reinforcement learning for wireless sensor localization,” in Advanced Information Net-
working and Applications Workshops (WAINA), 2012 26th International Conference on.
IEEE, 2012, pp. 760–765.
26. D. A. Tran and T. Nguyen, “Localization in wireless sensor networks based on support
vector machines,” IEEE Transactions on Parallel and Distributed Systems, vol. 19, no. 7,
pp. 981–994, 2008.
27. D. Janakiram, V. Reddy, and A. P. Kumar, “Outlier detection in wireless sensor networks
using bayesian belief networks,” in Communication System Software and Middleware,
2006. Comsware 2006. First International Conference on. IEEE, 2006, pp. 1–6.
28. J. W. Branch, C. Giannella, B. Szymanski, R. Wolff, and H. Kargupta, “In-network outlier
detection in wireless sensor networks,” Knowledge and information systems, vol. 34, no.
1, pp. 23–54, 2013.
29. S. Kaplantzis, A. Shilton, N. Mani, and Y. A. Sekercioglu, “Detecting selective forwarding
attacks in wireless sensor networks using support vector machines,” in Intelligent Sensors,
Sensor Networks and Information, 2007. ISSNIP 2007. 3rd International Conference on.
IEEE, 2007, pp. 335–340.
30. S. Rajasegarar, C. Leckie, M. Palaniswami, and J. C. Bezdek, “Quarter sphere based dis-
tributed anomaly detection in wireless sensor networks,” in Communications, 2007.
ICC’07. IEEE International Conference on. IEEE, 2007, pp. 3864–3869.
31. Z. Yang, N. Meratnia, and P. Havinga, “An online outlier detection technique for wireless
sensor networks using unsupervised quarter-sphere support vector machine,” in Intelligent
Sensors, Sensor Networks and Information Processing, 2008. ISSNIP 2008. International
Conference on. IEEE, 2008, pp. 151–156.
11
32. T. Avram, S. Oh, and S. Hariri, “Analyzing attacks in wireless ad hoc network with self -
organizing maps,” in Communication Networks and Services Research, 2007. CNSR’07.
Fifth Annual Conference on. IEEE, 2007, pp. 166–175.
33. A. Snow, P. Rastogi, and G. Weckman, “Assessing dependability of wireless networks u s-
ing neural networks,” in Military Communications Conference, 2005. MILCOM 2005.
IEEE. IEEE, 2005, pp. 2809–2815.
34. Y. Wang, M. Martonosi, and L.-S. Peh, “Predicting link quality using supervised learning
in wireless sensor networks,” ACM SIGMOBILE Mobile Computing and Communica-
tions Review, vol. 11, no. 3, pp. 71–83, 2007.
35. N. Ouferhat and A. Mellouk, “A qos scheduler packets for wireless sensor networks,” in
2007 IEEE/ACS International Conference on Computer Systems and Applications. IEEE,
2007, pp. 211–216.
36. Y. Gao, Y. Lin, and Y. Sun, “A wireless sensor network based on the novel concept of an
i-matrix to achieve high-precision lighting control,” Building and environment, vol. 70, pp.
223–231, 2013.
37. A. Mainwaring, D. Culler, J. Polastre, R. Szewczyk, and J. Anderson, “Wireless sensor
networks for habitat monitoring,” in Proceedings of the 1st ACM international workshop
on Wireless sensor networks and applications. Acm, 2002, pp. 88–97.
38. E. S. Nadimi, R. N. Jørgensen, V. Blanes-Vidal, and S. Christensen, “Monitoring and clas-
sifying animal behavior using zigbee-based mobile ad hoc wireless sensor networks and
artificial neural networks,” Computers and Electronics in Agriculture, vol. 82, pp. 44–54,
2012.
39. L. Paladina, A. Biundo, M. Scarpa, and A. Puliafito, “Self organizing maps for synchroni-
zation in wireless sensor networks,” in New Technologies, Mobility and Security, 2008.
NTMS’08. IEEE, 2008, pp. 1–6.
40. O. A. Postolache, J. D. Pereira, and P. S. Girao, “Smart sensors network for air quality
monitoring applications,” IEEE Transactions on Instrumentation and Measurement, vol.
58, no. 9, pp. 3253–3262, 2009.
41. K. Crammer, A. Kulesza, and M. Dredze, “Adaptive regularization of weight vectors,”
Machine learning, vol. 91, no. 2, pp. 155–187, 2013.
42. L. Yang, R. Jin, and J. Ye, “Online learning by ellipsoid method,” in Proceedings of the
26th Annual International Conference on Machine Learning. ACM, 2009, pp. 1153–1160.
43. J. Wang, P. Zhao, and S. C. Hoi, “Exact soft confidence-weighted learning,” arXiv pre-
print arXiv:1206.4612, 2012.
44. O. Younis and S. Fahmy, “Distributed clustering in ad-hoc sensor networks: A hybrid, en-
ergy-efficient approach,” in INFOCOM 2004. Twenty-third AnnualJoint Conference of the
IEEE Computer and Communications Societies, vol. 1. IEEE, 2004.
45. T. Zhang, R. Ramakrishnan, and M. Livny, “Birch: an efficient data clustering method for
very large databases,” in ACM Sigmod Record, vol. 25, no. 2. ACM, 1996, pp. 103–114.
46. S. Guha, R. Rastogi, and K. Shim, “Cure: an efficient clustering algorithm for large data-
bases,” in ACM Sigmod Record, vol. 27, no. 2. ACM, 1998, pp. 73–84.
47. N. Kimura and S. Latifi, “A survey on data compression in wireless sensor networks,” in
Information Technology: Coding and Computing, 2005. ITCC 2005. International Confer-
ence on, vol. 2. IEEE, 2005, pp. 8–13.
48. Y. Collet and M. Kucherawy, “Zstandard compression and the application/zstd media
type,” Internet Engineering Task Force, Tech. Rep., 2018. [Online]. Available:
https://tools.ietf.org/html/rfc8478
12
49. D. Sculley and C. E. Brodley, “Compression and machine learning: A new perspective on
feature space vectors,” in Data Compression Conference (DCC’06). IEEE, 2006, pp. 332–
341.
50. A. Adler, D. Boublil, and M. Zibulevsky, “Block-based compressed sensing of images via
deep learning,” in Multimedia Signal Processing (MMSP), 2017 IEEE 19th International
Workshop on. IEEE, 2017, pp. 1–6.