Research And Application On Feature Selection Based On Extending Of Rough Set

Rough set theory is developed on the basis of the classical set theory, and is a new mathematical theory for handling uncertainty and imprecision data. After years of development,it has found many theory and applications,especially the research and application of feature selection algorithm is one of the core research topics of rough set theory. Large number of studies found that the rough sets just works in discrete data, and the operation efficiency of the algorithm decreases with the increase of the attribute dimension. And how to find some extending of rough set that can deal directly with continuous data, and propose more effective and fast feature selection algorithms are difficulty of the field, but also a hot research direction.The core content of this article include:1.The improvement based on fuzzy rough set of feature selection algorithm. Algorithm initialized empty sets feature subset, successive to select the most important attributes to join feature subsets. In the search feature attribute, the attributes which importance were less than or equal to0degree were eliminated, reduced the number of calculation attribute importance and algorithms of running time.Finally, we did some experiments on three UCI datasets,and the experimental results showed the method in this paper was effective.2. Based on neighborhood rough set in feature select, a improvement is advanced to improve the neighborhood relationship.Because the delineation of the neighborhood did not well reflect differences in the strength of the distance between the different samples, the paper introduced the fuzzy membership function to the original model, so that made the neighborhood relationship of the sample fuzzy and retained information between the original data distance as much as possible. We also did some experiments on three UCI datasets,and the experimental results showed the method in this paper was effective.3. Combined with the feature extraction based on LE and the feature select based on extension of rough sets. Compared with the original method of breast X-ray image feature extraction and select based on PCA and classical rough sets, the method in this paper was effective by experimental results and analysis.