Penalised regression for high-dimensional data : an empirical investigation and improvements via ensemble learning