DESCRIPTION:
We develop predictive methods for the analysis of complex data. Since 1996, we have been working on biostatistical and environmental data, porting novel Neural and Statistical classifiers and regression models within Geographical Information Systems (GIS). Our concept is the application of machine learning on an integration of georeferenced data bases, digital maps, and landscape imagery (aerial and satellite remote sensing ), connecting predictive models with spatial DB. We also develop WEBGIS technologies for access to maps and databases, and for a fast geolocation of epidemiological data, allowing continuous update of the predictive models. To treat complex multidimensional problems, we study methods for model combination (local models, bagging and boosting of classification trees), data visualization and exploration (trellis graphics) , kernel methods (Support Vector Machines) and cost-sensitive methods. We develop statistical, pattern recognition and data management software, with applications for many Departments of the Province of Trento and for research in epidemiology. Predictive models for marketing applications have been developed for private clients. Since September 2004, our spin-off company MPA Solutions is managing maintainance and development of our software technologies. We support Open Source software development, for research and for innovation in the public administration, being the international headquarters of GIS GRASS and organizing events such as GFOSS.
|
