Enhanced IoT threat detection using Graph-Regularized neural networks optimized by Sea-Lion algorithm

Area of Science:

• Cyber Security
• Internet of Things (IoT)
• Machine Learning

Background:

• The interconnected nature of IoT systems introduces significant cyber security risks, including malware and software piracy.
• These threats can compromise sensitive data and damage organizational reputation.
• Existing methods for IoT threat detection require improvement to effectively mitigate these risks.

Purpose of the Study:

• To propose an advanced method for Internet of Things (IoT) threat detection.
• To enhance the accuracy and effectiveness of identifying both benign and malicious cyber security threats in IoT environments.
• To provide a robust solution for safeguarding sensitive data and protecting organizational reputation.

Main Methods:

• Utilized the Google Code Jam Dataset for analysis.
• Pre-processed data using Edge-Aware Smoothing Sharpening Filtering (EASSF) and feature extraction via General Synchro extracting Chirplet Transform (GSCT).
• Employed an AGRNN classifier optimized with the Sea-lion Optimization Algorithm for threat classification.

Main Results:

• Achieved significant accuracy improvements of up to 29.60% over existing methods for detecting benign threats.
• Demonstrated higher accuracy for malicious threat detection, with improvements up to 27.6%.
• Showcased superior F-measure and ROC values, confirming the method's effectiveness in cyber security threat detection.

Conclusions:

• The proposed IoT Threat Detection using Graph-Regularized Neural Networks Method offers a robust solution for enhancing IoT security.
• The approach effectively safeguards sensitive data and protects organizational reputation against cyber threats.
• This work provides a promising strategy for organizations aiming to strengthen their cyber security defenses.