Dynamic Latent Classification Model: Towards a More Expressive Model for Dynamic Classification

Monitoring a complex process often involves keeping an eye on hundreds or thousands of sensors to deter- mine whether or not the process is under control. We have been working with dynamic data from an oil pro- duction facility in the North sea, where unstable situa- tions should be identified as soon as possible. Motivated by this problem setting, we propose a generative model for dynamic classification in continuous domains. At each time point the model can be seen as combining a naive Bayes model with a mixture of factor analyzers (FA). The latent variables of the FA are used to capture the dynamics in the process as well as modeling dependences between attributes.