ABSTRACT
With the advancements in technology, a rapidly increasing number of devices are getting attached to the internet for their operation based on commands given by internet. These devices may include various home appliances like microwaves, electronic circuit boards, cameras, smart locks, smart toys, and other industrial equipment. This framework of devices connected to internet is called IoT. IoT framework is a combination of software, hardware, cloud services, internet, and user interfaces. Statistics shows that number of devices attached to IoT framework has increased drastically in recent times. The overall global market for IoT is expected to reach around $1.6 trillion by 2025 as per a forecasting report. Plenty of security techniques are available to for mobiles, computers, and laptops to safeguard against attacks, but not many options have been explored to provide sufficient safety to other devices. Many cases of breaching the security of cameras and other similar devices have been reported in past. As it is relatively a new field in this era of technology, it suffers from some of the major security-related issues.
Some issues include physical attacks, whereas some of them logically point to data when it is being gathered by various IoT objects. Major risks associated with IoT include unauthorized control and management of IoT-connected devices, alteration in data, distributed denial of service, Botnet attacks, duplication of devices, critical data revealing, spying and intruding through IoT devices, and many more. Approaches such as intrusion detection are recommended so as to mitigate or reduce the risk of data and components getting compromised. There has not been extensive research into finding anomalies in data collected by IoT components at application level. Hence, this research is focused on an attempt to bridge this gap. The motivation behind this research work is the prominent need of reliable and feasible solutions to protect IoT devices from attacks by detecting anomalies.
In this chapter, we are going to discuss various models based on machine learning to mitigate security concerns in IoT-connected devices. We have proposed a novel hybrid model based on partition around medoid (PAM) clustering and decision tree classifier. Classifier predicts its output as true based on its abnormal behaviour as shown by input variables. Its performance will be evaluated and compared with other existing machine learning techniques available for anomaly detection.
The proposed model has been tested and evaluated against various existing anomaly detection techniques for IoT. Its results are compared with contemporary strategies available to execute the same task. If the result is predicted as true value when some anomalies do exist in the applied data, then it is known as true positive (TP) value. If in some circumstances, a result is predicted as false even if there do exists some anomalies in the dataset, then it is known as false positive. Our proposed work aims to achieve better accuracy by reducing the number of false positive values. Results reveal that proposed model can detect anomalies in IoT-connected devices with higher accuracy.
