Self-moving Window with Long-Short Term Memory based Anomaly Detection in Edge-Assisted IoT Environments

Internet of Things (IoT) systems is widely used in many applications which generate huge amounts of sensory signals that involves to signify the state of systems. However, due to changes in IoT environment impact the detection of anomalies accurately which lead to risk of losing information. To overcome this limitation, a Self-moving Window with Long Short-Term Memory (SW-LSTM) is proposed for accurate detection of anomalies. The self-moving window improved the ability of LSTM to identify subtle anomalies that evolve over time, leading to more accurate detection of unusual patterns or device behavior in IoT environments. Initially, the data about attacks and malicious nodes are acquired from the UNSW-15 dataset and preprocessed by score normalization technique. Then, the optimal set of features are selected by Particle Swarm Optimization (PSO) algorithm. Finally, the proposed SW-LSTM model detects the anomalies in edge assisted IoT network effectively. The experimental results of proposed method achieved accuracy of 93.4% which is greater when compared to existing detection approaches such as Recurrent Neural Network (RNN), Convolutional Neural Network (CNN).