A Hybrid Approach: Utilising Kmeans Clustering and Naive Bayes for IoT Anomaly Detection

TL;DR

This paper proposes a hybrid anomaly detection algorithm for IoT devices combining k-means clustering and Naive Bayes, achieving high accuracy and scalability for diverse IoT environments.

Contribution

It introduces a novel hybrid method that integrates unsupervised and supervised learning to improve anomaly detection across multiple IoT devices.

Findings

Detection accuracy ranges from 90% to 100%.

The algorithm demonstrates high precision and recall.

It offers a scalable solution for IoT anomaly detection.

Abstract

The proliferation and variety of Internet of Things devices means that they have increasingly become a viable target for malicious users. This has created a need for anomaly detection algorithms that can work across multiple devices. This thesis suggests a potential alternative to the current anomaly detection algorithms to be implemented within IoT systems that can be applied across different types of devices. This algorithm is comprised of both unsupverised and supervised machine areas of machine learning combining the strongest facet of each. The algorithm involves the initial k-means clustering of attacks and assigns them to clusters. Next, the clusters are then used by the AdaBoosted Naive Bayes supervised learning algorithm in order to teach itself which piece of data should be clustered to which specific attack. This increases the accuracy of the proposed algorithm by adding clustered data before the final classification step, ensuring a more accurate algorithm. The correct indentification percentage scores for this proposed algorithm range anywhere from 90% to 100%, as well as rating the proposed algorithms accuracy, precision and recall. These high scores achieve an accurate, flexible, scalable, optimised algorithm that could potentially be in different IoT devices, ensuring strong data integrity and privacy.