Identification of Faulty Sensors and Predicting Missing Sensor Data in IoT-Enabled Smart Buildings

Smart buildings rely on continuous sensor data to support energy efficiency, occupant safety, and automated control. However, faulty or missing signals not only undermine reliability in Internet of Things (IoT) environments but also weaken the operational resilience of wireless infrastructures by creating opportunities for service disruption and adversarial exploitation. This study proposes an integrated Artificial Neural Network (ANN) framework that combines Convolutional Neural Network-based fault detection with Long Short-Term Memory autoencoder-based reconstruction. Data were collected from DHT22 temperature and humidity sensors, pre-processed through cleaning, fault simulation, and sliding-window segmentation, and then used to train lightweight models suitable for realtime edge deployment. The evaluation results show that the fault detection module achieved 93.5 % accuracy on test data, with strong precision-recall balance and an AUC of 0.95, ensuring reliable identification of sensor malfunctions while limiting false alarms. The reconstruction module produced accurate recovery of corrupted signals, with an overall RMSE of 0.1181 and$R^{2}=0.9365$, closely aligning reconstructed outputs with ground truth. By unifying detection and recovery in a single lowlatency pipeline, this work advances the robustness, security, and continuity of wireless-enabled building management systems. The findings contribute to a practical, resource-efficient approach for enhancing the fault-tolerance and operational resilience of IoT infrastructures.