ALERT-IoT: Advanced Anomaly Detection Framework for IoT Environment Using Deep Learning

Abstract

Anomaly detection entails identifying and flagging uncommon or irregular activities within a network or system. It assists in pinpointing any deviations or unexpected patterns in the data generated by interconnected devices. Current techniques frequently have difficulties precisely identifying abnormalities in the midst of enormous data sets and various attack patterns. In this paper, a novel AnomaLy dEtection using attention based BiGRU DenseNet in IoT (ALERT-IoT) technique has been proposed for effective anomaly detection in IoT environments, which combines Autoregressive Density Estimation (ADE) and Jensen-Shannon Divergence (JSD) with attention based BiGRU-DenseNet for effective anomaly detection. By integrating data preprocessing, feature extraction using the Butterfly Optimization Algorithm, and robust anomaly score calculation, the proposed method demonstrates superior performance compared to existing Improved Bacterial Foraging Optimization with Optimal Deep Learning for Anomaly Detection (IBFO-ODLAD), Tiny Anomaly Detection (TinyAD) and Adaptive Anomaly Detection Approach Toward Concept Drift (ADTCD) techniques. Extensive evaluation on IoTID20 and Voraus-AD datasets showcases the superiority of the proposed ALERT-IoT method over existing techniques. The proposed ALERT-IoT method is calculated using numerous assessment metrics, including accuracy, recall, precision, and detection rate. The detection rate of the proposed ALERT-IoT method is 25.49%, 27.37%, and 5.60% higher than the existing IBFO-ODLAD, ADTCD, and TinyAD, techniques respectively.