Most of the research activity carried by Ro.S.A. is to reshape modern statistical technique with major emphasis on robustness. Robust statistics is a branch for which results from a method are not disturbed by change of assumptions or presence of outliers. In other words, when a robust method is applied the presence of influential observations have little or no effect on the final results. As a byproduct, extreme cases are highlighted and might be analysed further, as sometimes they are the very important. Ro.S.A. is currently leading in many specific statistical methods used for Big Data with a novel approach which is robust, dynamic and fully tailored for automated machine learning.

Data are everywhere. IBM estimates that several million of gigabyte are created every year, with a yearly average growth of 50%. In Ro.S.A. we are committed to propose tools for such a challenge.

OUTLIERS AND ERRORS

Handling outliers does not simply imply to trash them in the bin. Outliers must be managed with the same care that is put to all data. Errors are inconsistent values which are not compatible with a given set of hypnotises and, when present, they have not to influence the performance, but once highlighted, are no longer relevant.

OUTLIERS AND PATTERN

Standard outdated approaches of pre-cleaning of data is very much arbitrary and often based on a variable-by-variable approach. Beside of being time consuming, this might also break multivariate latent pattern that can be the most important structure for a given problem! In Ro.S.A. we tackle this issue with new fast approaches which limit the arbitrariness.

BIG DATA

Big Data is a keyword that define a huge amount of data with or without a structure. Businessmen face daily the problem of tacking actions based on decisions data driven. Robust methods for big data provide invaluable support to take better actions.

AUTOMATION

The FSDA Toolbox, jointly developed with the Joint Research Centre of the European Commission, is designed for an automated implementation of robust algorithms for many statistical analysis. Ro.S.A. is continuously developing codes that might be used within third parties software for full customisation to any need. Automation is meant by an automated process, model and data driven, with no arbitrary choices.

Research topics