ENHANCING CYBERSECURITY WITH AI-DRIVEN DYNAMIC HONEYPOTS

Abstract

Cybersecurity threats have evolved, requiring advanced defense mechanisms to detect and mitigate attacks effectively. Honeypots serve as decoy systems to attract and analyze malicious activities, but traditional honeypots lack adaptability and rely on static rule-based detection. This paper explores the integration of Artificial Intelligence (AI) and Machine Learning (ML) in honeypots to enhance threat detection and dynamic response mechanisms. We compare Traditional Rule-Based Detection, Random Forest, and XGBoost models to classify attacks based on honeypot log data. Additionally, an AI-Driven decision-making layer is implemented to dynamically respond to threats by categorizing attack severity and selecting appropriate countermeasures. Due to a lack of real-world attack data, publicly available honeypot logs from Kaggle were used for training and evaluation. The result demonstrates that XGBoost outperforms other models, achieving higher accuracy and recall in detecting malicious activity. While the current system operates based on predefined AI rules, future enhancements could incorporate real-time adaptive honeypots capable of modifying network defenses dynamically based on attack pattern