Early stopping aggregation in selective variable selection ensembles for high-dimensional linear regression models.

•A novel algorithm is devised to get more effective variable selection ensembles.•ST2E is extended to high-dimensional cases by an effective prescreening step.•A subensemble is attained by sorting members and then fusing some top-ranked ones.•The accuracy-diversity patterns of both full and pruned subensembles are studied.•Higher ranking and selection accuracies are gained by a pruned ensemble.