Ensemble Feature Weighting Based on Local Learning and Diversity

Recently, besides the performance, stability (robustness, i.e., variation in feature selection results due to small changes data set) of is received more attention. Ensemble where multiple outputs are combined yield robust without sacrificing performance an effective method for stable selection. In order make further improvements (classification accuracy), diversity regularized ensemble weighting framework presented, which base selector based on local learning with logistic loss its robustness huge irrelevant features and samples. At same time, sample complexity proposed algorithm analyzed VC-theory. The experiments different kinds sets show that can achieve higher accuracy than other ones strategy (such as weighting)