Порівняльний аналіз ефективності LSTM та Bi-LSTM мереж у задачах детектування аномалій в IoT-трафіку

Abstract

The paper addresses the problem of securing Internet of Things (IoT) networks by detecting anomalous activity in traffic. A comparative analysis of two deep learning architectures, unidirectional (LSTM) and bidirectional (Bi-LSTM) recurrent neural networks, is performed. The ability of the models to account for temporal dependencies in network packet sequences for cyberattack identification is investigated. The analysis results demonstrate the advantages of using BiLSTM in terms of detection accuracy for complex anomaly patterns due to data processing in both forward and backward directions.