Research On Feature Selection Method Based On Dominant Neighborhood Rough Se

In the era of big data,the rapid development of information technology creates a boom that all kinds of industries are flooded with digital data in various forms.Massive information waits to be explored from those complex data.However,characterized by large scale,multi-modality and realtime growth,big data brings about severe challenges to knowledge discovery and modeling.And that is the reason why it is necessary to accelerate the advancement of data mining,which helps find out mass of valuable knowledge.Nowadays,granular computing has become more and more important in intelligent computing.As an emerging theory to process information of uncertainty,it simulates the way that human cognize and granulates the whole information into multiple simpler sub-blocks.Then complex tasks can be learned from the viewpoints of multi-level,multi-perspective and multi-granulation.Considering that neighborhood relation confines to select required samples since there may also be partial relations even among similar objects.Set in numerical ordered data,this thesis aims to select key features and furthermore obtain valuable knowledge from tremendous data.Connected with methods of granular computing such as rough set and fuzzy set,the thesis intends to improve the dominance-based neighborhood rough set from three aspects.Meanwhile,both static and incremental algorithms are designed followed with numerical tests on UCI datasets to verify the effectiveness and efficiency of them.The main innovations of this thesis are as follows:1.The classical models ignore the various features’ significance to decision and sets strict limits through dominance relations.Then we design a heuristic feature selection approach and its relative incremental learning mechanism based on the weighted dominance-based neighborhood rough set(WDNRS).At first,corresponding weights are learned from the data and dominance tolerance is introduced to construct WDNRS.Then condition entropy in matrix form is calculated through dominance matrix and recognized as evaluation metric.Finally,numerical experiments are carried out to verify that it is effective and efficient for the designed method to select features in dynamic datasets when objects increase.2.The neighborhood rough set is vulnerable to misclassify due to the random selection of neighborhood radius,lack of the upper approximation in decision metric and the vulnerability to noise.Grounded on intuitionistic fuzzy ordered information system(IFOIS),we come up with a scored intuitionistic fuzzy dominance-based neighborhood rough set(SIFDNRS)using relative decision self-information to select features.Initially,IFOIS is constructed via score function.Then we adjust the chosen radius by neighborhood surrounding function to avoid the influence of data distribution.And the scored IFDNRS is obtained.Based on the relative decision self-information,a heuristic is proposed and results from UCI tests show that the proposed method outperforms comparative approaches.