TeVAE: A Variational Autoencoder Approach for Discrete Online Anomaly Detection in Variable-state Multivariate Time-series Data

The introduction of TeVAE marks a significant advancement in the field of anomaly detection, particularly in the context of automotive testing where the complexity of data is on the rise. As manual evaluation methods reach their limits, the need for automated online anomaly detection becomes critical. TeVAE, a temporal variational autoencoder, addresses this need by effectively detecting anomalies with a minimal false positive rate of only 6%, while successfully identifying 65% of actual anomalies. This model not only avoids the bypass phenomenon but also introduces a novel method for remapping individual data windows to a continuous time series, enhancing its applicability in real-world scenarios. The ability of TeVAE to perform well with smaller training datasets further underscores its potential as a robust tool in the realm of data analysis, paving the way for more efficient and reliable anomaly detection systems.