Lightweight Internet of Things Botnet Detection Using One-Class Classification.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• The proliferation of Internet of Things (IoT) devices has led to increased vulnerability to security threats.
• IoT botnets are increasingly used to launch distributed denial of service (DDoS) attacks, disrupting online services.
• Early detection of IoT botnet infections is crucial for mitigating damage and maintaining service availability.

Purpose of the Study:

• To propose a novel machine learning solution for the early detection of IoT botnets.
• To develop a lightweight and accurate detection model suitable for resource-constrained IoT environments.
• To address the challenge of detecting botnets in heterogeneous network settings.

Main Methods:

• Development of a one-class classifier based on the k-Nearest Neighbors (KNN) algorithm for anomaly detection.
• Implementation of a feature selection strategy using filter and wrapper methods to optimize model performance.
• Evaluation of the proposed model using diverse datasets from various network scenarios.

Main Results:

• The proposed one-class KNN classifier demonstrated high accuracy in detecting IoT botnets at early infection stages.
• The machine learning model proved to be a lightweight solution, efficient in feature selection and detection.
• Consistent performance improvements were observed across three different evaluation datasets.

Conclusions:

• The developed machine learning approach offers an effective and efficient method for early IoT botnet detection.
• The one-class classifier provides a robust solution for identifying malicious activities in heterogeneous IoT environments.
• This research contributes to enhancing the security and resilience of IoT ecosystems against botnet threats.