Development of a Robust Intrusion Detection System Using Attention Mechanism for Imbalanced Network Traffic in IoT

Abstract

The Internet of Things (IoT) is ever-changing, which is reshaping industries and societies. It involves connecting devices and data to enable automation and consolidation of processes. This technological shift brings about significant changes in business operations and societal interactions. With progress in IoT, there is a greater need to address security issues. Significant threats to complex resources and company operations are associated with unauthorized access which may lead to system outrage. This study predicts network traffic patterns within IoT devices using a Long Short- Term Memory (LSTM) model integrated with an attention mechanism purposely aimed at detecting intrusion while enhancing security consciousness across various networks in the IoT. Hence, we have used UNSW-NB15 data set for this study. The results indicate that LSTM-based attention system achieves 99% accuracy identifying binary data and multiclass classification 97% across the entire data set. By concentrating on the top ten features, a classification accuracy of 97% for binary classification and 96% for multi-class was achieved. Such results imply that analyzing IoT network traffic could be done using the LSTM-based attention model. With this knowledge we can design safer and more stable networks between multiple IoT devices.