Application of Information Theoretical Models for AI-Driven Cybersecurity Systems
Manas Kumar Yogi, Yamuna Mundru, Atti Manga Devi · Advances in computational intelligence and robotics book series · 2024
The integration of AI-driven models into cybersecurity has significantly advanced threat detection and mitigation. This study enhances AI system performance using information theoretical models by incorporating entropy and mutual information to improve accuracy and robustness in identifying complex threats. Entropy measures data uncertainty, while mutual information quantifies information exchange between variables, refining AI detection capabilities. We compare Logistic Regression and Random Forest classifiers using precision, recall, accuracy, and F1 score metrics. Results show that models using information theoretical features perform better, reducing false positives and identifying more true threats. This is crucial in preventing alert fatigue and ensuring reliable cybersecurity. Advanced feature integration makes AI models more adept at recognizing patterns and anomalies, enhancing threat detection, especially against complex and subtle threats. Ultimately, this research works will propel the future cyber security designers with holistic perspectives.