R. Blanquero, E. Carrizosa, P. Ramírez-Cobo, M. R. Sillero Denamiel
Lasso has become a benchmark data analysis procedure, and it has been deeply studied and extended by many authors. In particular, we can find in the recent literature some Lasso variants which deal with data collected from distinct strata, as it is standard in many biomedical contexts. In this work we propose a novel variant of Lasso in which the degrees of reliability of the different data sources are taken into account. However, this is not enough to guarantee the good performance across all the considered sources. Therefore, we shall additionally demand that overall performance measures, as well as performance measures for the data from each source attain certain threshold values. As a result, a generalized regression model is obtained, addressed by solving a nonlinear optimization problem.
Palabras clave / Keywords: model selection, Lasso, heterogeneous data analysis, performance constraints
Programado
Sesión GT02-2: Análisis Multivariante y Clasificación (AMyC-2). Organizadora: Eva Boj del Val
29 de mayo de 2018 12:20
Sala 5
