Sensor Alert Verification for Incident Operational Response (SAVIOR)

We are developing a decision support architecture for chemical sensing networks, with the goal of improving their capability by interpreting alerts in the context of other available information. Most simply, health and status metrics derived from sensor data might indicate that false alerts are more than usually likely. In principle, however contextual information can come from a vast number of sources, including field or laboratory sensor calibration data and prior characterization of potential interferants, as well as site-specific data such as the results of environmental samples, meteorological data, historical alert incidence, and observations from on-site observers. To accommodate this diversity of data sources, we apply Bayesian network techniques that are commonly employed for information management in many expert systems, including for automated medical diagnosis and target identification in various military applications.