‘Cost-sensitive classification. A mathematical optimization approach’

Rafael Blanquero Bravo, Alba Victoria Olivares Nadal and Vanesa Guerrero Lozano, University of Seville; Alfredo Marín Pérez, University of Murcia; Josefa Ramírez Cobo, University of Cádiz; María Dolores Romero Morales, Copenhagen Business School; Belén Martín Barragán, The University of Edinburgh; and María Remedios Sillero Denamiel, Universal Dx.

The aim of this 24-month project is to develop new supervised classification models which take into account different performance measures, namely, the misclassification rates and predictive values of the different classes, as well as measurement costs.

To address this cutting-edge challenge, a cohesive research team will use its solid expertise in Mathematical Optimization to express the construction of the classifier as an optimization problem. This way, an overall performance measure will be optimized under constraints, which control the achievement in the different individual performance measures under consideration. The models will be validated in problems of two domains in which the team has extensive expertise, namely, medical diagnosis (cancer diagnosis through tumor markers) and credit scoring.

With this project, advances on the state of the art will be obtained at both a methodological level (powerful and versatile classifiers are to be designed) and a practical level, since we will give answer to challenges posed in the two applications fields addressed.