The new baseline for high dimensional dataset by ranked mutual information features

Feature selection is a process of selecting group relevant features by removing unnecessary for use in constructing the predictive model. However, high dimensional data increases difficulty feature due to curse dimensionality. From past research, performance model always compared with existing results. When attempting new dataset, current practice benchmark dataset obtained including all features, redundant and noise. Here we propose optimal baseline mean ranked using mutual information score. The quality depends on contained more contains better number achieve this will be at same time, serve as guideline needed method. We also show some experimental results that proposed method provides fewer features.