Introducing Machine Learning to Wireless Sensor Networks: Requirements and Applications

Introducing Machine Learning to Wireless Sensor Networks: Requirements and Applications

Vidit Gulyani (HMR Institute of Technology and Management, Delhi, India), Tushar Dhiman (HMR Institute of Technology and Management, Delhi, India) and Bharat Bhushan (HMR Institute of Technology and Management, Delhi, India)
DOI: 10.4018/978-1-7998-5068-7.ch001
OnDemand PDF Download:
No Current Special Offers


From its advent in mid-20th century, machine learning constantly improves the user experience of existing systems. It can be used in almost every field such as weather, sports, business, IoT, medical care, etc. Wireless sensor networks are often placed in hostile environments to observe change in surroundings. Since these communicate wirelessly, many problems such as localisation of nodes, security of data being routed create barriers for proper functioning of system. Extending the horizon of machine learning to WSN creates wonders and adds credibility to the system. This chapter aids to present various machine learning aspects applied on wireless sensor networks and the benefits and drawbacks of applying machine learning to WSN. It also describes various data aggregation and clustering techniques that aim to reduce power consumption and ensure confidentiality, authentication, integrity, and availability amongst sensor nodes. This could contribute to design and alter pre-existing ML algorithms to improve overall performance of wireless sensor networks.
Chapter Preview


Cluster of sensor nodes deployed to observe different environment scenarios by sharing sensed information to other nodes and central body called base station (BS) collectively attribute to Wireless sensor networks. These nodes operate on low power and are stationed, specifically to observe change in surroundings. Sensors have limited resource and are stationary. These are posted in various scenarios to sense and report any change in environment. WSN’s are used in field of healthcare, hostile and military grounds, weather forecasting etc.

Sensor nodes observe numerous contextual changes such as pressure, weather, thermal and optical. These are arranged in a specific fashion. Sensor nodes report change in event of their interests. If more than one node in the network sense the same event, they all work in collaboration by gathering data from the event (Lan et. al, 2017). Only one node is allowed to generate the final report. Report is delivered to Base station either directly or by hopping the information on other sensor nodes through wireless medium. Base station unit is a gateway between nodes and external world. Each node in the network has following three components:

  • a)

    Sensor subsystem: senses any change in environment.

  • b)

    Processing subsystem: it generates report of the change recorded from environment

  • c)

    Communication system: this unit sends the report to base station and receive any instruction from base station.

Figure 1 depicts a block diagram for Wireless sensor Networks.

Figure 1.

A typical representation of Wireless sensor networks


Requirements for WSN mainly comprise non-functional requirements of the system. Major non-functional requirements are security, less power consumption, reliability of the network, fault and intrusion detection. Inducing Machine Learning algorithms to WSN increases reliability as any malicious activity such as tampering of sensor data, and intrusions are easily detected and removed from network by clustering the previous data and removing outliers from each cluster (Zhao et al, 2015; Wang et al, 2019).

Machine learning emerged in mid-20th century to serve as ground for Artificial Intelligence (AI). Increased use of software’s worldwide generated enormous data in clouds, hence machine learning expanded its horizons to almost every field. Different researchers define ML in varied tones as per the application ground. Combining all these definitions, machine learning can be defined as: Training a machine model to predict future outcomes by analysing old patterns and learn new ways of behaviour using statistical data as experience. Machine learning improves experience of prebuild software systems. E.g. spam detection in Gmail, AI based opponent players in games. Voice assistants and many more. Machine learning also serves as a ground for deep learning systems such as autonomous vehicles, self-trained robots etc. With robust algorithms ML analyses previously occurred patterns in various fields e.g. business, sports, weather predictions to predict future outcomes with high confidence level .

Coupling WSN and ML adds credibility to network. Ad-hoc networks have increased the use of Machine Learning algorithms from past few decades. Non-functional requirements in WSN are attained by applying machine learning algorithms on data generated by sensor networks some of which are listed below:

  • 1.

    WSN are mostly positioned by armed forces in hostile environments to keep an eye on enemy’s activities. It becomes important to ensure security as the system can be attacked and data can be modified. ML algorithms play their parts by recognizing any change in behaviour of nodes by using k-NN and other clustering algorithms. Details about these algorithms are discussed in Section III.

  • 2.

    Furthermore, Intrusion and anomalies in the network are eliminated. It improves QoS (Quality of Service) thereby increasing reliability on the network.

  • 3.

    Sensor networks are resource limited. Various clustering algorithms use very less computational power and proves to be an asset for wireless sensor networks.

Complete Chapter List

Search this Book: