LSTM-Autoencoder-Based Anomaly Detection for Indoor Air Quality Time-Series Data

Anomaly detection for indoor air quality (IAQ) data has become an important area of research as the is closely related to human health and well-being. However, traditional statistics shallow machine-learning (ML)-based approaches in anomaly IAQ could not detect anomalies involving observation correlations across several points (i.e., often referred long-term dependencies). We propose a hybrid deep-learning model that combines long short-term memory (LSTM) with autoencoder (AE) tasks address this issue. In our approach, LSTM network comprised multiple cells work each other learn dependencies time-series sequence. The AE identifies optimal threshold based on reconstruction loss rates evaluated every all sequences. Our experimental results, Dunedin carbon dioxide (CO2) dataset obtained through real-world deployment schools New Zealand, demonstrate very high robust accuracy rate (99.50%) outperforms similar models.