Reliable identification of chemical hazards in multidimensional datasets is a key challenge for the deployment of effective chemical vapor detection systems. Much of the work to date has focused on, and is applicable only to, individual technologies. As new techniques are developed, many of the same data processing challenges recur. In Phase I of this project, we will conduct a feasibility study and develop a proof-of-concept software system that is applicable to multiple detection technologies and has the capability to detect low-level signals from generalized inputs. Our system is novel in that it uses a data fusion framework in which individual features are annotated with statistical measures or probabilities, and the results of various runs are combined statistically. Toward this end, we will translate several algorithms that have been successful in bioinformatic sequence analysis to the chemical analysis domain, extending them as necessary to handle richer data sets. In a Phase II continuation, we will develop a full beta version of the software and analyze data from several different technologies
