Attribute Reduction Based on Lift and Random Sampling

As one of the key topics in development neighborhood rough set, attribute reduction has attracted extensive attentions because its practicability and interpretability for dimension or feature selection. Although random sampling strategy been introduced to avoid overfitting, uncontrollable may still affect efficiency search reduct. By utilizing inherent characteristics each label, Multi-label learning with Label specIfic FeaTures (Lift) algorithm can improve performance mathematical modeling. Therefore, here, it is attempted use Lift guide reduce uncontrollability sampling. In this paper, an based on called ARLRS proposed, which aims searching Firstly, used choose samples from dataset as members first group, then reduct group calculated. Secondly, divide rest into groups have symmetry structure. Finally, reducts are calculated group-by-group, guided by maintenance reducts’ classification performance. Comparing other 5 strategies set theory over 17 University California Irvine (UCI) datasets, experimental results show that: (1) significantly time consumption reduct; (2) derived provide satisfying tasks.