Graph Neural Network-Based Adaptive Data Imputation and Outlier Detection for Streaming IoT Time Series

The proliferation of Internet of Things (IoT) ecosystems has led to an explosion of streaming time series data from a multitude of sensors. This data is often incomplete due to sensor failures or transmission errors, and can contain anomalous readings due to noise or malicious activity. Traditional data imputation and outlier detection methods are often ill-suited for the dynamic and non-stationary nature of streaming IoT data, and fail to capture the complex spatial and temporal correlations between interconnected sensors. To address these challenges, this paper introduces a novel framework that leverages Graph Neural Networks (GNNs) for adaptive online learning to perform real-time data imputation and outlier detection. Our approach models the IoT sensor network as a dynamic graph, where GNNs are used to learn the intricate dependencies between sensors and infer missing values and identify anomalies by considering both the local neighborhood and the global network structure. A key innovation of this work is the development of an adaptive learning mechanism that continuously updates the model to handle non-stationary data streams and concept drift, enabling it to maintain high performance over extended periods without the need for periodic retraining. Extensive experiments on real-world streaming IoT datasets demonstrate that our GNN-based framework significantly outperforms state-of-the-art baselines in both imputation accuracy and anomaly detection, while maintaining low latency suitable for edge deployment.