Anomaly Detection with SDAE
This work addresses data quality issues in building energy management, but it is incremental as it applies existing autoencoder methods to a specific dataset.
The paper tackled anomaly detection in building energy data by comparing Simple, Deep, and Supervised Deep Autoencoders, finding that the Supervised Deep Autoencoder detected the most anomalies in test datasets.
Anomaly detection is a prominent data preprocessing step in learning applications for correction and/or removal of faulty data. Automating this data type with the use of autoencoders could increase the quality of the dataset by isolating anomalies that were missed through manual or basic statistical analysis. A Simple, Deep, and Supervised Deep Autoencoder were trained and compared for anomaly detection over the ASHRAE building energy dataset. Given the restricted parameters under which the models were trained, the Deep Autoencoder perfoms the best, however, the Supervised Deep Autoencoder outperforms the other models in total anomalies detected when considerations for the test datasets are given.