Reinforcement Q-Learning-Based Adaptive Encryption Model for Cyberthreat Mitigation in Wireless Sensor Networks.

Area of Science:

• Computer Science
• Cybersecurity
• Network Engineering

Background:

• Wireless Sensor Networks (WSNs) face increasing cyber threats, demanding adaptive security solutions.
• Resource constraints in WSNs limit the implementation of static, high-level encryption.
• Existing security mechanisms often lack the adaptability to handle dynamic threat landscapes.

Purpose of the Study:

• To propose a reinforcement learning-based adaptive encryption framework for WSNs.
• To dynamically scale encryption levels based on real-time network conditions and threat classification.
• To optimize the trade-off between energy efficiency and security robustness in WSNs.

Main Methods:

• A deep learning-based anomaly detection system for threat classification (low, moderate, high).
• Integration of dynamic Q-learning and double Q-learning for adaptive security policies.
• Formulation as a Markov Decision Process (MDP) with a tailored reward function.
• Implementation of an ϵ-greedy exploration-exploitation mechanism and dynamic hyperparameter tuning.

Main Results:

• Achieved a 30.5% reduction in energy consumption.
• Maintained a 92.5% packet delivery ratio (PDR).
• Demonstrated 94% mitigation efficiency against cyberattacks (DDoS, black-hole, data injection).
• Reduced latency by 37% compared to conventional encryption.

Conclusions:

• The reinforcement learning-driven adaptive encryption framework is effective and scalable for resource-constrained WSNs.
• The proposed system balances energy efficiency and security robustness dynamically.
• The framework shows significant improvements in performance metrics and attack mitigation, suitable for IoT applications.