Skip to content
Open access

H-RADIS: An Adaptive Edge-Centric Lightweight Framework for Real-Time Multi-Stream Anomaly Detection in Smart Building IoT Systems

Sep 2026 · Proceedings of the Pakistan Academy of Sciences: A. Physical and Computational Sciences · 0 citations

Abstract

The rapid adoption of smart, real-time, QoS-aware IoT applications generates a massive amount of multi-stream data collected from heterogeneous interconnected devices. Extracting valuable and immediate insights from this data requires real-time processing and analytics. Real-time anomaly detection enhances these insights by identifying deviations in data as they occur, ensuring operational reliability, energy efficiency, and resilience against faults. Detecting anomalies in such streams involves analyzing individual streams and identifying complex interrelationships among multiple streams, where irregular patterns may emerge collectively or over time. Therefore, this paper proposes H-RADIS, an adaptive Half-Space Trees (HST) based framework specifically designed for real-time, multi-stream anomaly detection on edge devices. It effectively detects point and combined anomalies and employs Complex Event Processing (CEP) to detect collective anomalies. The framework integrates a sliding window-based dynamic threshold with adaptive parameter tuning, enabling the H-RADIS to cope with varying situations in real time more effectively. The performance of H-RADIS is evaluated against RIVER-based online machine learning algorithms, Local Outlier Factor (LOF), and OneClassSVM (OCSVM) across two different datasets by augmenting them with synthetic anomalies. The experimental results indicate that H-RADIS consistently improves anomaly detection performance over the collective LOF–OCSVM baseline, with relative gains in F1-score (26.08%), Recall (23.88%), Precision (22.22%), and ROC-AUC (17.69%). In terms of computational efficiency, H-RADIS achieves a relative 48.65% increase in throughput, together with relative reductions of 34.71% in processing time, 34.69% in energy consumption, and 0.21% in memory usage, while maintaining comparable CPU utilization. Overall, the comparative results indicate that, under the evaluated experimental conditions, H-RADIS is a resource-efficient approach for real-time anomaly detection in smart-building IoT environments.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.