New article related to IDUNN!! A novel edge architecture and solution for detecting concept drift in smart environments

Advances in AI and IoT Optimize Smart City Applications within European IDUNN Cybersecurity Project

The expansion of the Internet of Things (IoT) and artificial intelligence (AI), along with the adoption of 5G technology and progress towards 6G, have led to the generation of vast amounts of data in smart cities and buildings. Despite this, much of the data remains underutilized. A significant challenge identified is concept drift, where the statistical properties of real-world data streams change over time due to unforeseen factors, leading to less efficient predictive models.

Addressing this issue, recent research conducted under the European IDUNN cybersecurity project proposes a new computing architecture. This architecture is designed for various smart city applications within edge micro data centers (EMDC) across a hybrid cloud–edge continuum, optimizing the deployment of AI workloads. A novel, feedback-driven automated concept drift detection and adaptation methodology integrates long short-term memory (LSTM) models with techniques like the Page–Hinkley test (PHT), adaptive windowing (ADWIN), and the Kolmogorov–Smirnov windowing (KSWIN).

The methodology has been tested using real-world data streams from environmental sensors at the University of Oulu Smart Campus, initially validated with synthetic datasets to identify known concept drift points. The implementation demonstrated significant improvements, reducing the Mean Absolute Percentage Error (MAPE) from 8.5% to 3.88% upon applying concept drift detection. This research not only tackles the challenges faced in predictive modeling but also enhances the efficiency of smart city applications, showcasing the effectiveness of the solutions provided by the IDUNN project.

Link to the article (Open Access)

Authors: Hassan Mehmood, Ahmed Khalid, Panos Kostakos, Ekaterina Gilman, Susanna Pirttikangas