| Feature selection is an important strategy for knowledge reduction of rough sets.Interval-valued data,as an extension of single values,can better represent uncertain information from the perspective of uncertainty measures.However,in real-world applications,the feature values in interval-valued data change over time.For dynamic interval-valued data,it is very time-consuming to select feature subsets using existing methods because they require recalculating interval-valued data from scratch when feature values change.For this reason,we investigate incremental methods for feature selection in dynamic interval-valued data environments,which can select new feature subsets based on previous results.The corresponding incremental update mechanisms are introduced according to different dynamic changes,and finally,by comparing the feature subset selection results of the incremental and non-incremental algorithms on a common data set,it can be concluded that the two proposed incremental algorithms are more efficient,especially when there are multiple objects changing feature values simultaneously,and the incremental feature selection algorithm proposed in the paper achieves a satisfactory results.The main research work in this paper is as follows.Conditional entropy is constructed.In view of this,based on two kinds of feature value change problems that arise in interval-valued decision systems,i.e.,dynamic change of feature values in a single object and dynamic change of feature values under multiple objects at the same time,the θ-similarity class is first constructed using the similarity of interval-valued data,and then the incremental update strategy is designed based on the improved θ-conditional entropy,jointly with the incremental feature selection algorithm based on the improved θ-conditional entropy,and then designed an incremental feature selection algorithm based on the improvedθ-conditional entropy and joint local update strategy.Finally,a series of static and dynamic comparison experiments are conducted on the publicly available UCI dataset,and the final experimental results verify the effectiveness and efficiency of the proposed algorithm,and the incremental feature selection algorithm under multiple feature value changes is more superior.(1)In order to address the dynamic update of feature values of objects,the update of local feature values are investigated and an incremental update mechanism based on θ-conditional entropy is constructed.In view of this,based on two kinds of feature value change problems that arise in interval-valued decision systems,dynamic change of eigenvalues of a single object and dynamic change of eigenvalues under multiple objects at the same time,the theta-similarity class is first constructed using the similarity of interval-valued data,and then the incremental update strategy is designed based on the improved θ-conditional entropy,jointly with the incremental feature selection algorithm based on the improved θ-conditional entropy,and then designed an incremental feature selection algorithm based on the improvedθ-conditional entropy and joint local update strategy.Finally,a series of static and dynamic comparison experiments are conducted on the publicly available UCI dataset,and the final experimental results verify the effectiveness and efficiency of the proposed algorithm,and the incremental feature selection algorithm under multiple feature value changes is more superior.(2)For the dynamic changes of feature sets in interval-valued decision systems,the changes of local feature sets are studied and the incremental update mechanism based on θ-discernibility degree is constructed.In view of this,the incremental feature selection algorithm is designed by first analyzing the local update mechanism based on θ-discernibility degree in the case of feature set addition and deletion,respectively,according to the dynamic change problem,i.e.,feature set addition and deletion that arises in interval-valued decision systems,and then jointly designing the incremental feature selection algorithm by forward search strategy.Finally,a series of comparative experiments are conducted to validate the proposed algorithm with four other different feature selection algorithms on the publicly available UCI dataset,and the final experimental results demonstrate the outstanding advantages of the proposed algorithm in terms of enhancement efficiency. |
