Fusing binary and continuous output of multiple classifiers

This paper describes a fusion architecture and implementation for classifiers with binary as well: as continuous output. The fusion scheme is represented as a multi-layered architecture which structures the approach into pre-processing, analysis, and post-processing. All components are partitioned into layers where each layer performs logical tasks. In particular, the pre-processing component is partitioned into a temporal layer and a scaling layer. The analysis component is partitioned into strengthening and weakening layers. The post-processing component is structured into suppression and exception handling layers. Manipulations within these layers transform the binary classifier output as well as continuous output into a single continuous domain. Because the fusion architecture is designed in a modular fashion,. additional modules can be relatively easily. be added or removed. The modular design also allows re-use of the core fusion engine for other domains. We show results of this architecture applied to a system monitoring environment of industrial equipment.