IDENTIFICATION D’ANOMALIES PAR APPRENTISSAGE SUPERVISÉ A L’AIDE D’UNE ARCHITECTURE PROFONDE HYBRIDE ORIENTÉE SIMILIARITÉ : UNE APPROCHE POUR LA GESTION DES CLASSES DÉSÉQUILIBRÉES.

Anomaly detection in cybersecurity is a major challenge due to the significant imbalance between normal data, which are largely predominant, and attacks, which, although rare, are highly critical. To address this issue, we propose a supervised learning approach based on a similarity-oriented hybrid deep architecture, using a Siamese network that combines CNNs for feature extraction and LSTMs for modeling temporal dependencies. This architecture aims to improve the discrimination of minority classes while enhancing the model’s robustness against adversarial attacks. Experimental results show that the Siamese CNN–LSTM model achieves an accuracy of 98.9%, highlighting its effectiveness in handling class imbalance and improving the performance of intrusion detection systems.