Exploring Artificial Intelligence and Machine Learning Approaches With Large Language Models In Cybersecurity Contexts

The objective of this work is to explore the relation between typical Artificial Intelligence (AI) and Machine Learning (ML) algorithms and the potentiality of Large Language Models (LLMs) to classify and predict in the context cyberse-curity environments, in particular, in network-based anomaly detectors in user profiles, benign and malicious traffic flows, including Denial of Service (DoS), large-scale attacks in Internet-of-Things (IoT) environment, malware memory analysis attack, network intrusion detection systems and in internet of medical things security solutions. First, using open-source tools for AI/ML we explore exhaustively four benchmarking datasets and create several classification mod-els for each one using Decision Trees (DT), Random Forest (RF), Gradient boosting and others. We evaluate the models using standard metrics, ensuring high accuracy relative to the specificity of each dataset. These models are typi-cally used to classify and predict new cases. However, recognizing the efficien-cy of these algorithms in handling tabular data and for predicting as well as the potentialities of LLMS, we propose to convert a DT and RF tree in rules and knowledge from the AI/ML models into prompts and feed them into the LLMs, aiming to enhance their performance to classify and predict new cases We com-pare and achieve high performance in terms of accuracy using the AI/ML ap-proaches with the ChatGPT, Gemini and Ollama. With this hybrid view and the promising results of both approaches we create an instrumental tool to elucidate the predictions to prevent new cases in the context of cybersecurity environ-ments. To foster further research in this context the datasets, code, benchmark, and pretrained models are available on the GitHub open-source repository.