Résumé:
The rapid growth of the internet in recent years has made cybersecurity a significant challenge. The traditional and standard Intrusion Detection Systems (IDS) which work based on known attack patterns are not effective enough and not sufficient to detect modern threats nowadays. For this reason, in this project, we aimed to enhance the functionality of IDS using either Machine Learning (ML) or Deep Learning (DL) to detect attacks. To reach our goal, we compared several models to decide which one is the best and gives best performance .However, to ensure that individuals' data stay safe, we adopted Federated Learning (FL), which enables the model to learn from different distributed data sources and devices without sharing private data. We evaluated our work using a real- world dataset UNSW-NB15, we implemented both a Federated MLP and a Federated Random Forest (RF) that returned best results among Ml and DL algorithms, using different aggregation strategies. Our final federated MLP model achieved over 98% across accuracy, precision, recall, and F1-score, proving that federated deep learning can deliver state-of-the-art results while preserving data confidentiality.
Keywords: Cybersecurity, Intrusion Detection System (IDS), Machine Learning (ML), Deep Learning (DL), Federated Learning (FL), Multi-Layer Perceptron (MLP), Random Forest (RF).
