Real data sets from a wide variety of fields violate the idealized assumptions inherent in standard statistical theory. Robust data analysis methodology aims to mitigate the impact of such violations. Robust methods are usually developed to handle multivariate data. However, monitoring studies often contain information such as functional, set-valued, or different kinds of incomplete data. Robust methods for these complex data types are scarce and involve critical computational challenges. New models, methods and efficient, numerically stable, and well-conditioned robust strategies are essential to improve knowledge extraction from non-perfect and non-standard datasets. Applications include the analysis of climate data, medical monitoring and diagnosis, trading and financial forecasts. The aim is to create an interactive network spanning computing, statistics, machine learning, and mathematics with the necessary expertise required to develop such strategies in close collaboration with end-users. Software and guidelines will be developed. The Action will provide European scientists with cutting-edge data analysis tools which will be suitably disseminated by disparate means such as training schools, conferences and publications. Improved decision-making tools for preventing-mitigating policies will be derived. Thus, scientific, technological and social challenges will be tackled by the creation of a proper framework to coordinate and optimize research efforts.
Robust methods - large non-perfect and non-standard datasets - numerical estimation - combinatorial optimization - parallel implementation