Data mining for diagnosis and prognosis

The DISCO team has for many years acquired experience in the field of data mining, classification methods, information fusion based on learning fundamentals, integrating fuzzy logic concepts, with present works along two directions :

  • Imprecise and uncertain data processing: a new type of data, interval-valued, has been introduced as an information source, enabling thus processing of imprecise information without any prior transformation, simultaneously with quantitative and qualitative type data.



 

  • Feature and sensor selection: a selection method has been developed in order to determine the most relevant informative features, i.e. enabling the optimization of diagnosis/prognosis performances.

    These generic developments have been successfully applied, in particular in the medical domain (diagnosis and extraction of new signatures for breast cancer prognosis) in collaboration with Institut Claudius Regaud at Toulouse and in the field of chemical processes. Studies are presently performed on the selection of a set of sensors which guaranty the diagnosticability and the selection of control loops with the aim of improving the controllability.