Partly adaptive elastic net and its application to microarray classification

This paper presents a new extension of the elastic net for simultaneous gene selection and microarray classification. By introducing the proper data-driven weights to the penalty terms, the partly adaptive elastic net is proposed, which can encourage an adaptive grouping effect and reduce the influence of the wrong initial estimation. A fast-solving algorithm is also developed in the line of pathwise coordinate descent. Experiments performed on the two cancer data sets are provided to verify the obtained results.