Emergent Deep Learning for Anomaly Detection in Internet of Everything
Youcef Djenouri, Djamel Djenouri, Asma Belhadi, Gautam Srivastava, Jerry Chun‐Wei Lin
Abstract
This research presents a new generic deep learning (DL) framework for anomaly detection in the Internet of Everything (IoE). It combines decomposition methods, deep neural networks, and evolutionary computation to better detect outliers in IoE environments. The data set is first decomposed into clusters, while similar observations in the same cluster are grouped. Five clustering algorithms were used for this purpose. The generated clusters are then trained using DL architectures. In this context, we propose a new recurrent neural network for training time-series data. Two evolutionary computational algorithms are also proposed: 1) the genetic and 2) the bee swarm, to fine-tune the training step. These algorithms consider the hyperparameters of the trained models and try to find the optimal values. The proposed solutions have been experimentally evaluated for two use cases: 1) road traffic outlier detection and 2) network intrusion detection. The results show the advantages of the proposed solutions and a clear superiority compared to state-of-the-art approaches.