A Hybrid AI-Based Model for Cyberattack Detection and Mitigation

Calle Huaman D.G.; Dextre Alarcón J.S. · 35th International Conference on Computer Theory and Applications, ICCTA 2025 - Conference Proceedings (2025) · DOI: 10.1109/iccta68914.2025.11520037

Resumen

In recent years, information security has gained critical importance, driven by an alarming surge in cyberattacks. In Peru alone, more than 15 billion attempts have been reported - a trend that continues to rise. Enterprise networks are among the most affected, particularly in critical sectors such as healthcare, where the protection of information is essential. In this context, a hybrid AI-based detection and mitigation model is proposed that is designed to deliver a lightweight, efficient, and low-cost solution capable of automatically identifying and responding to threats in real time without compromising network performance. Network traffic is analyzed on a Raspberry Pi 5 using Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTMs), with tools such as TensorFlow/Keras, Flask, SQLite, and iptables. The architecture is tailored for resource-constrained environments, facilitating deployment in real-world networks. The methodology was structured into five phases: tool selection, preparation of the CICIDS2017 dataset, model training, integration in a controlled environment, and validation at a hospital in Lima. An accuracy of 97% was achieved, providing a solid foundation for future real-world deployments.

Instituciones: Universidad Peruana de Ciencias Aplicadas