This work has opened the way for several future avenues of research that we are currently investigating. First, a popular alternative to LDA in classification problems is logistic regression. It has been recently motivated by ROC considerations (McIntosh and Pepe [19]). While it is possible to formulate a LASSO estimation for logistic regression models, adapting the LASSO-SVM equivalence to this situation requires new algorithms. It will also be important to compare the performance of the two L1-regularized procedures (LDA and logistic regression) on real and simulated microarray datasets.
