improv data mine util project sampl mark perform data mine process depend induc knowledg cost process cost acquir pre process train exampl cpu cost model induct cost commit error recent progress sampl strategi maxim data mine util propos strategi base repeat acquisit addit train exampl util decreas observ paper present altern project sampl strategi fit function partial learn curv partial run time curv obtain small subset potenti data project function analyt estim optim train set size propos approach evalu varieti benchmark dataset rapidmin environ machin learn data mine process result show learn run time curv project data point lead cheaper data mine process common progress sampl method 