Missing data imputation remains a persistent challenge in Internet of Things and sensor-based environments, particularly for applications related to smart cities and urban data analysis. In this study, we compare imputation methods for univariate urban temperature time series collected from three distinct weather stations with significant missing segments. To evaluate the performance of each method, artificial gaps of varying lengths are systematically introduced into the data. We assess the accuracy of eXtreme Gradient Boosting (XGBoost) regression in comparison with more conventional techniques such as spline interpolation. While XGBoost performs well for short and medium gaps, its effectiveness diminishes for longer gaps due to its limited capacity to capture the cyclic patterns inherent in temperature data. To address this, we propose a context-aware Bidirectional Long Short-Term Memory network that captures both past and future temporal dependencies. The model performs multi-step predictions using a one-day context window, avoiding the cumulative error typically propagated in recursive single-step approaches. We also compared the performance of our proposed model with Multi-layer Perceptron. Our results show that the Bidirectional Long Short-Term Memory model performs better than MLP overall and significantly outperforms XGBoost, especially for long missing blocks. These findings demonstrate the effectiveness of context-aware deep learning—particularly Bidirectional Long Short-Term Memory model for robust imputation of cyclic temperature time series in sensor networks.

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.