Community-Based Model Sharing and Generalisation: Anomaly Detection in IoT Temperature Sensor Networks

TL;DR

The paper addresses anomaly detection in large-scale IoT temperature sensor networks with heterogeneous deployments. It proposes a Community of Interest (CoI) framework that clusters sensors using a fused similarity matrix derived from temporal Spearman correlations, spatial Gaussian distance decay, and elevation similarity, and trains representative-station autoencoders (MLP/LSTM/BiLSTM) optimized by Bayesian methods ($s_{ij} = \exp(- d_{ij}^2 / (2 \sigma^2))$ and $s_{ij}^{(e)} = \exp(- (d_{ij}^{(e)})^2 / (2 \sigma^2))$). The approach demonstrates strong within-cluster anomaly detection and selective cross-cluster generalisability, with cluster 2 showing weaker transfer. This framework reduces computational overhead by sharing models per community, enabling scalable anomaly detection in IoT deployments, and points to future edge deployment, region-based distribution, and XAI directions for deeper interpretability.

Abstract

The rapid deployment of Internet of Things (IoT) devices has led to large-scale sensor networks that monitor environmental and urban phenomena in real time. Communities of Interest (CoIs) provide a promising paradigm for organising heterogeneous IoT sensor networks by grouping devices with similar operational and environmental characteristics. This work presents an anomaly detection framework based on the CoI paradigm by grouping sensors into communities using a fused similarity matrix that incorporates temporal correlations via Spearman coefficients, spatial proximity using Gaussian distance decay, and elevation similarities. For each community, representative stations based on the best silhouette are selected and three autoencoder architectures (BiLSTM, LSTM, and MLP) are trained using Bayesian hyperparameter optimization with expanding window cross-validation and tested on stations from the same cluster and the best representative stations of other clusters. The models are trained on normal temperature patterns of the data and anomalies are detected through reconstruction error analysis. Experimental results show a robust within-community performance across the evaluated configurations, while variations across communities are observed. Overall, the results support the applicability of community-based model sharing in reducing computational overhead and to analyse model generalisability across IoT sensor networks.

Figures (9)

• Figure 1: Schematic overview of community-based anomaly detection framework
• Figure 2: Location of meteorological stations in the sensor network.
• Figure 3: Average daily temperature patterns across the selected meteorological stations.
• Figure 4: Distribution of daily temperature values across stations represented as boxplots.
• Figure 5: Monthly temperature distributions for a subset of representative stations.