A comparison of two strategies for scaling up instance selection in huge datasets