International Conference on Innovative Advancement in Engineering and Technology (IAET-2020)

IoT and Machine Learning- A Technological Combination for Smart Application

Dr.Jyoti Batra Arora*

Project Associate,ICAI,Noida, India, [email protected]

ABSTRACT

With the rapid developments in hardware, software and communication technologies, a new era of internet connected sensory devices has begun
which provides a link between physical and cyber worlds. It is estimated that by 2020 internet connected devices being used will be between 25 and 50
billion. Internet connected devices referred as Internet of Things. IoT provides a huge range of services and applications covering all the domains ranging
from infra-structure, retail, transportation and personal health care. Fughua etal. (2015) poised that IoT has empowered the development in internet
protocols and communication technologies. In IoT applications, intelligent processing and analysis of big data act as key for their development. Data
science technologies help to find new pattern and new insights from data to make IoT applications more intelligent. Data science with IoT is mainly used
in various sectors dealing with volume, velocity and pattern recognition. Because of predictive analysis of machine learning, software becomes able to
predict incoming desired and undesired events. Therefore, machine learning system is not only recognizing the abnormal behaviour but also help to
understand and establish long term trends.A continue correction and supervision is required for effectiveness and efficiency in data analysis. This paper is
having two sections- pertaining the different applications of IoT where machine learning helps to make the intelligent system and futuristic scope of IoT
and machine learning and how these two technologies help in development of communication devices. This paper will provide the answers to questions
such as what is the taxonomy of machine learning that can be adopted in IoT, how can machine learning be applied to IoT applications.

Keywords:Machine Learning, IoT, Data analysis, Data Science.

1. INTRODUCTION

IoT is defined as the network of physical objects empowered with limited computation, storage and communication capabilities
and embedded with electronics, software and network connectivity. Novo, Bijar and Ocak (2015) defined IOT as an
interconnected and distributed network of embedded system of communicating through wireless or wired communication
technologies. IoT network brings challenges in maintenance of devices, if these devices are prone to, storage, communication and
privacy. Ammar etal.(2018) discussed about the amount of research going on architecture, communication and computation,
security and privacy of IoT .

IoT devices generate massive amount of data which can be used further patterns, predictions and assessments. This data creates a
different front for the current data processing mechanism. A new mechanism is required to harness the value of IoT generated
data. The most suitable computational paradigm would be Machine Learning (ML).Machine learning provides embedded
intelligence to IoT Devices and infer the knowledge from device generated data. M. at. E, (2018) poised that machine learning
enhances the ability of smart devices to vary or automate the situation or behaviour based on the knowledge. ML can be used in
classification, regression and density estimation. ML techniques and algorithm can be leveraged in IoT based applications such as
fraud detection, malware detection and speech recognition and provide intelligent services.

2. SECURITY CHALLENGES IN IoT APPLICATION

IoT has great influence by creating a new dimension in the internet world. The main issues in IoT services and application are
security and privacy. Security of IoT device may include architectural security, data security, communication security, malware
analysis and so on. Fernandes etal.(2017) poised the similarities and dissimilarities of the security issues in IoT and traditional IT
devices. They stated that the most common driving factors are- software, hardware, network and applications. A cross layer
design and optimized algorithm are required as the solution to security and privacy issues of IoT. IoT devices further may require
a new breed of cryptographic and other algorithm to cope up with security and privacy. However, increase in number of IoT
devices may increase the challenges in security mechanism. A holistic security and privacy approach is more appreciated than

February21-22,2020/ Jaipur NationalUniversity, Jaipur, India Page 1

Electronic copy available at: https://ssrn.com/abstract=3548431

 International Conference on Innovative Advancement in Engineering and Technology (IAET-2020)
existing security solution. This new approach is to provide new intelligent, robust, evolutionary and scalable mechanism to
control the challenges and insecurity in IoT applications.

ML is one of those intelligent methods that provide the optimal solution through learning using past experience or example data.
ML uses mathematical techniques to build models of behaviours. ML can also enable the smart devices to learn without use of
explicitly programming. ML has multidisciplinary nature that may include artificial Intelligence, optimization theory and
cognitive science. This feature allows ML to be useful where human expertise cannot be used such as robotics and speech
recognition and real life problem. It provides the solution to IoT where solution to a specific problem changes in time. Though
ML technique is very reliable but sometimes it produces false positive and true negative results so it requires proper guidance and
modification. An advance version of ML that is Deep Learning (DL) helps to overcome this issue and determines the accuracy of
prediction by itself. Mohammadi etal. (2018) and Nguyen etal. (2017) found DL is more suitable for prediction and classification
task in innovative IOT application because of its self-service nature.

IoT network produces a huge amount of data which is very useful for ML and DL. From many researches, it has been observed
that these both can be used in IoT network for analysis of security, attack detection and prevention and malware analysis. S.
etal.(2018) discussed the challenges related to deployment of these models on resource- constrained IoT devices. These
challenges occur as it is essential to reduce the processing and storage overhead of IoT devices. IoT devices are resource
constrained so it is not feasible to use sophisticated security mechanism against malicious attacks. The pervasive deployment of
large number of devices also increases the attack surface. IoT devices uses various communication technologies such as zigbee,
LoWPAN, z-Wavw and NFC. Al-sarawi (2017) stated the limitations of these communication technologies from security point of
view. Apart from aforementioned challenges, other challenges such as scalability, complexity, addressing and in sufficient
resource utilizations are also related to IoT (Liu(2018), Chen (2016)).

3. EXISTING MACHINE LEARNING BASED SOLUTION TO PROVIDE SECURITY IN IoT

Various machine learning algorithm such as–supervised learning, unsupervised learning, semi supervised and reinforcement
learning algorithm can be used in IoT security. Supervised learning is used where specific targets are achieved from certain set of
inputs. It is performed when type of input data and desired output are known. It can be used for spectrum sensing, channel
estimation, adaptive filtering, security and localization problems. Unsupervised learning does not require labelled data as the
environment only provides input without desired targets. It is used in anomaly, fault and intrusion detection, cell clustering and
load balancing. Reinforcement Learning (RL) can be performed where no specific outcomes are defined and agents learnt from
feedback after interacting with environment. Supervised and unsupervised learning techniques basically focus on data analysis
whereas reinforcement learning is used in comparison and decision making problems. It is used to train the system where system
has to discover the structure within raw data. Semi supervised learning which exists between supervised and unsupervised
learning can be used where labels in most of the observation are not present. DL is most suitable for distributed computing,
learning and Analysis of sheer amount of unlabelled, uncategorised and unsupervised data.

L. etal.(2015) Poised that DL is originated from ANN, so learning is followed by the iterative adjustment of weights among each
pair of neurons. Wang etal. (2017) stated that DL can be used in various application of ML such as speech recognition, computer
vision and NLP through improved classification modelling. It basically works on function approximation, estimation and the
learning capabilities and provides solution to IOT especially in security and privacy issue. IoT devices are resource constraint so
may not be capable to handle complex computational algorithm. Thus DL algorithm show better performance with lower latency
and complexity compared to conventional theories and techniques (Wang etal., 2017).

Traditional RL is not capable enough therefore a combination of DL and RL is required to find the best policy and quality of an
action in a given state (Mahmud etal.,2017). RL and DL mutually benefits each other. Though DL is capable of learning from
complex patterns, yet also prone to mis-classification (Mohammadi etal.,2018). S.etal.(2017) poised that DRL as the combination
of RL and DL which integrates decision making of RL and perception of DL. D.etal.(2017) stated the uses of DRL in an
application developed by google i.e. “ AlphaGo”. DRL algorithm is generally used for security and DDOS detection in IoT
network. A scheme is proposed for dynamic IoT water marking to identify the cyber attacks by Ferdowsi etal.(2017).

Most of the IoT applications are based on data exchange across different platforms. The data retrieved from IoT applications are
processed and passed through a decision-support system to make sense out of it. The data flow to these processes is same, but

February21-22,2020/ Jaipur NationalUniversity, Jaipur, India Page 2

Electronic copy available at: https://ssrn.com/abstract=3548431

 International Conference on Innovative Advancement in Engineering and Technology (IAET-2020)
processes may differ depending upon architecture of IoT. Whenever any user requires data, the authentication of user or
application is required. Authentication is the basic security requirement else it will deny the access request. Network access
controls are also required which is also very challenging. To grant and revoke the access to specific user to critical data set of IoT,
ML based access control mechanism is used.

4. LIMITATION OF USING MACHINE LEARNING IN IoT

There are certain inherent uncertainties associated with IoT data which require considerable modification however ML
technologies are not efficient enough for this (Qiu,2016). Therefore, a few limitations are also associated while using ML in IoT.
Algorithm of machine learning possesses issues with memory, and computational complexity. ML techniques are only limited to
low dimensional problem as they lack in scalability. ML works on constant stream of data in real time so sometimes it is not
suitable for smart IoT devices which work on real time data processing. LHeureux (2017) discussed that how predictive ability of
an algorithm decreases with the increase in the dimensionality of data. IoT network generates diversify data which differ in
semantics and format, hence exhibit syntactic and semantic heterogeneity which raises a problematic issue to ML (Bogale,2018).

In the case of ML statistical properties of entire dataset remain the same therefore, in real time applications where data from
various sources have been different in formatting and representations causes problems for ML algorithms as these algorithms are
not efficient to work on semantic and syntactic diversified data. Merging of ML algorithm with existing streaming solutions
enhances the overall complexity of an algorithm.

5. Futuristic Scope of IoT and Machine Learning

ML is the driving for artificial intelligence. The major advantages of using ML system are heuristic learning, decision tree for
fundamental administration and data acquisition (Mandal etal.2012). All data science based applications like data mining,
information retrieval system, search engine and big data analysis use ML algorithms. It also help to find applications in computer
vision for object identification (Salau,2018). IoT is the most used and recent application of ML. Many researchers presented the
survey of using IoT with ML in different applications and services. Kumar etal.(2019) presented the survey of IoT and its
application in built engineering. Buczak and Guven (2016) has given the survey on IoT in cyber security system using data
mining techniques. Most of the proposed schemes focus on technical area not on uses needs. Though ML helps to reduce burden
of user in their daily life by providing them technological advantages, yet raises many critical issues. In future ML with
combination of IoT needs to work on several critical issues such as scalability, cost, battery of sensors, handling of multiple
sensors, time elapse and many more.

Conclusion

ML techniques are being used to make IoT devices and applications more smart and intelligent. It helps to reduce computational
cost, enhanced computing power and integration of various technological breakthroughs have made this possible. ML creates
model that design, test, and train the data sets.ML algorithm help to detect pattern and similarities in data set, yet the fundamental
limitation of is that, it mostly needs dataset to learn from, and then the learned model is applied to the real data. DL help to
overcome the limitations of ML and used in industry domain. Apple’s Siri, Microsoft’s Cortana, Amazon’s Alexa and Google
Photo are the examples of DL algorithm. DL,RL and DRL are the prominent research area to learn automated extraction from
large amounts of high dimensional unsupervised data. ML and DL are assumed as efficient techniques for classification and
predictions in many applications though these may not able to cater all challenges faced by IoT network.

REFERENCES

1. O. Novo, N. Beijar, and M. Ocak, “Capillary Networks - Bridging the Cellular and loT Words ,” IEEE World Forum on Internet of Things (WF-IoT),
vol. 1, pp. 571–578, December 2015.
2. A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and M. Ayyash,(2015) ,Internet of things: A survey on enabling technologies, protocols, and
applications, IEEE Communications Surveys Tutorials, 17, 2347–2376.
3. M. Ammar, G. Russello, and B. Crispo,(2018).Internet of things: A survey on the security of IoT frameworks, Journal of Information Security and
Applications, 38, 8 – 27.

February21-22,2020/ Jaipur NationalUniversity, Jaipur, India Page 3

Electronic copy available at: https://ssrn.com/abstract=3548431

 International Conference on Innovative Advancement in Engineering and Technology (IAET-2020)
4. M. at. el,(2018),Machine Learning for Internet of Things Data Analysis:A Survey , Journal of Digital Communications and Networks, Elsevier, 11–56.
5. E. Fernandes, A. Rahmati, K. Eykholt, and A. Prakash,(2017),Internet of things security research: A rehash of old ideas or new intellectual
challenges?,IEEE Security Privacy, 15(4)79–84.
6. S. et al.,(2018),Deep Learning for the Internet of Things, IEEE Journal of Computer, vol. 51, 32–41.
7. S. Al-Sarawi, M. Anbar, K. Alieyan, and M. Alzubaidi,(2017). Internet of things (IoT) communication protocols: Review, in 2017 8th International
Conference on Information Technology (ICIT), 685–690.
8. Y. Liu, Y. Kuang, Y. Xiao, and G. Xu,(2018). Sdn-based data transfer security for internet of things, IEEE Internet of Things Journal, vol. 5, 257– 268.
9. J. Chen, S. Li, H. Yu, Y. Zhang, D. Raychaudhuri, R. Ravindran, H. Gao, L. Dong, G. Wang, and H. Liu,(2016). Exploiting icn for realizing service-
oriented communication in IoT, IEEE Communications Magazine, vol. 54, pp. 24–30.
10. L. et al.,(2015). DeepEar: robust smartphone audio sensing in unconstrained acoustic environments using deep learning, ACM International Conference
on Pervasive and Ubiquitous Computing, vol. 1, pp. 283–294.
11. T. Wang, C.-K. Wen, H. Wang, F. Gao, T. Jiang, and S. Jin,(2017),Deep Learning for Wireless Physical Layer:Opportunities and Challenges, IEEE
China Communication, vol. 14, pp. 92–111.
12. M. Mohammadi, A. Al-Fuqaha, M. Guizani, and J. Oh,(2018). Semisupervised deep reinforcement learning in support of IoT and smart city services,
IEEE Internet of Things Journal, vol. 5, pp. 624–635.
13. N. D. Nguyen, T. Nguyen, and S. Nahavandi,(2017).System design perspective for human-level agents using deep reinforcement learning: A survey,
IEEE Access, vol. 5, pp. 27091–27102.
14. A.Ferdowsi and W.Saad.(2017),Deep Learning based Dynamic Water Marking for Secure Signal Authentication in the Internet of Things.
15. A. LHeureux, K. Grolinger, H. F. Elyamany, and M. A. M. Capretz.(2017).Machine Learning With Big Data: Challenges and Approaches. IEEE Access,
vol. 5, pp. 7776 – 7797.
16. J. Qiu, Q. Wu, G. Ding, Y. Xu, and S. Feng (2016), “A survey of machine learning for big data processing,” EURASIP Journal of Advance Signal
Process, May 2016.
17. T. E. Bogale, X. Wang, and L. B. Le,(2018). Machine Intelligence Techniques forNext-GenerationContext-AwareWirelessNetworks, Arxiv,vol.19, pp.
1–10.
18. Mandal, I., and Sairam, N.(2012) Accurate prediction of coronary artery disease using reliable diagnosis system, Journal of Medical Systems, 36(5),
3353-3373.
19. Salau, A. O.(2018). Development of a vehicle plate number localization technique using computer vision, PhD Thesis, Obafemi Awolowo University,
Ile-Ife, Nigeria, 200.
20. Kumar, A., Salau, A.O., Gupta, S. and Paliwal, K.(2019). Recent trends in IoT and its requisition with IoT built engineering: A review. In: Rawat, B.,
Trivedi, A., Manhas, S., Karwal, V. (eds) Advances in Signal Processing and Communication. Lecture Notes in Electrical Engineering, 526, 15-25.
Springer Singapore. DOI: 10.1007/978-981-13-2553-3_2
21. Buczak, A. L. and Guven, E(2016). A survey of data mining and machine learning methods for cyber security intrusion detection, IEEE
Communications Surveys and Tutorials, 18(2), 1153-1176, 2016.
22. M. Mahmud, M. S. Kaiser, A. Hussain, and S. Vassanelli,(2018). Applications of deep learning and reinforcement learning to biological data. IEEE
Transactions on Neural Networks and Learning Systems, vol. 29, pp. 2063–2079.
23. D. et al.,(2018). Review on the Research and Practice of Deep Learning and Reinforcement Learning in Smart Grids, CSEE Journal of Power and
Energy Systems, vol. 4, pp. 362–370.

February21-22,2020/ Jaipur NationalUniversity, Jaipur, India Page 4

Electronic copy available at: https://ssrn.com/abstract=3548431