A Light-Weight Real-Time Anomaly Detection Framework for Edge Computing

Abstract

In the Internet of Things IoT environment, the quality of delivered services essentially relies on several factors, such as the quality of data collected by the numerous numbers of embedded sensors, and the quality of the underlying network. Hence, a fault or anomaly arising from the hardware, software, or network can have devastating consequences regarding the overall quality of the corresponding services. Since that anomalous behaviour can be existed independently due to either an attack or sensor malfunction, developing methodologies that can detect anomalies and identify their sources seamlessly in real-time is a crucial demand to provide robust and reliable IoT services. This research investigates the potential data quality degradation caused by anomalies through analyzing sensory-related data. The ultimate aim is to utilize unsupervised deep-learning techniques, namely: AE-LSTM, and VAE-LSTM and adopt the edge computing concept that employs edge devices to detect anomalous data, describe, and analyze the effect of such anomalies from the quality provided perspective. This is achieved by introducing a Light-Weight Real-Time Anomaly Detection Framework that comprises two distinct layers: a back layer which includes a deep-learning-based anomaly detection trainer, and a front layer which is an edge device that acts as a Real-time anomaly detector. The evaluated models showed outperformed results compared to the state-of-the-art One-Class Support Vector Machine OCSVM unsupervised learning technique with up to 95% F1 score in detecting anomalies imposed into sensors’ readings.