Invasive Weed Optimization Based Weighted Fuzzy-rough Feature Selection

Feature selection is a dimensionality reduction technique with the purpose of selecting the most predictive subset from those input features.This is a problem encountered in many areas of computational intelligence.The benefits of feature selection include simplifying feature internal-relationship and improving prediction performance.Now,there are a lot of algorithms of feature selection based on fuzzy-rough set research.For example,the Genetic algorithm,the Ant Colony Optimization(ACO)and the Particle Swarm Optimization(PSO),and so on.These algorithms are excellent in robustness and solving problem ability.These methods only select the best individuals from the attribute set,however,some individuals with lower original fitness may have important information.Therefore,these algorithms may lead to important information loss.This paper studies characteristics of Invasive Weed Optimization(IWO).We found that characteristics of IWO can make the fuzzy-rough feature selection more comprehensive.The core of IWO is that some weeds with lower fitness may have important information.IWO gives those individuals with lower fitness some opportunities to breeding.The offspring individuals are randomly spread around their parents individuals according to a Gaussian distribution.The algorithm maintains the diversity of the species in the early and middle iterations,and it ensures the global optimal solution in final iterations.Therefore,this paper proposed the weighted fuzzy-rough feature selection algorithm based on Invasive Weed Optimization(IWO-FRFS)that combines the characteristics of IWO and fuzzy-rough set theory.We verify the feature selection results using the fast attribute reduction algorithm based on fuzzy-rough set.Finally,we put the algorithm in fourteen benchmark data sets and four mammographic data sets with realistic background,and we make comparison between our feature selection results and the results of other algorithms(ACO-FRFS and PSO-FRFS)from the classification accuracy and the AUC.The data analysis shows that most feature selection results of IWO-FRFS can represent the original data sets well,and the overall performance of IWO-FRFS are better than the ACO-FRFS and the PSO-FRFS.Meanwhile,mammographic risk analysis is an important task for assessing the likelihood of women developing breast cancer.Therefore,it also proves that IWO-FRFS is a meaningful research.