Research On Multi-category Classification Based On Sequential Three-way Decision

In the decision-making problem,considering that the accept decision and reject decision in the two-way decision can make great costs or consequences,in order to avoid the risk of direct acceptance or rejection,the three-way decision introduces the third decision-making behavior:delay decision.When the information is insufficient,the deterministic decision cannot be made immediately.The delay decision is made to reduce the loss caused by the wrong acceptance or the wrong rejection to a large extent,which is superior to the two-way decision.Three-way decision is one of the core ideas of decision-theoretic rough sets.It contains three kinds of decision-making behaviors,namely,accept decision,reject decision and delay decision.When the decision object is divided into the positive region,it means that the object makes the accept decision,the object divided into the negative region represents the reject decision,and the object divided into the boundary region represents the delay decision,which are giving semantic interpretation for positive region,negative region and boundary region of rough set.However,most of research and applications regarding the three-way decision are oriented to the two-category problem and are not applicable to the actual decision-making issue.For example,when a patient goes to a hospital for treatment,it is not only necessary to judge whether the patient has a cold or not,but also to determine which kind of disease the patient is suffering from,such as viral hepatitis,pneumonia or other diseases.In addition,when the object makes a delay decision,it needs to further collect the state information of the object,so that the object can make a deterministic decision,which constitutes sequential three-way decision.For the problem of multi-category classification decision-making,this thesis takes the sequential three-way decision as the research background and carries out the following research:(1)In order to solve the multi-category classification decision problem,three multi-category classification models based on sequential three-way decision are proposed.Firstly,with the multi-category decision information system as the research object,three calculation methods for transforming multi-category classification into multiple two-category classifications are given.In the first method,the m-category classification is transformed into m two-category classifications,that is,a certain class and all the remaining classes are formed into a two-category.Therefore,three-way decision of the m two-category classifications are respectively made;The second method divides the m-category into two multi-category,and then makes three-way decision based on two multi-category to obtain positive region,negative region,and boundary region.For positive region and negative region,if they still contain multi-category,the method continues to process the positive region and negative region separately until the number of categories of objects in all positive region and negative region becomes one;In the third method,two of the m categories are composed into two-category and processed separately.Therefore,the m(m-1)times of three-way decision can be made for m decision classes.Secondly,considering that the sequential three-way decision utilizes multi-level granular structure,the cost of decision-making can be effectively reduced.This thesis uses sequential three-way decision to implement the above three multi-category classification methods.Therefore,this thesis constructs a multi-level granular structure of three-way decision based on the attribute sequence.Then,on each attribute level,three-way decision can be made based on the above three multi-category classification methods.For the object that cannot determine the decision class,the decision will be made again at the next level,thus forming three multi-category classification models based on sequential three-way decision.The calculation process of the three models is illustrated by the examples.Finally,the effectiveness of the proposed models is verified by simulation experiments.(2)For the change of objects in multi-category classification decision problems,a multi-category classification increment algorithm based on cost-sensitive sequential three-way decision is proposed.First,for each level of the sequential three-way decision,the loss function matrix is defined separately.By using the method of m-category classification is transformed into m two-category classifications,the threshold of each decision class is calculated to make three-way decision.For the objects that cannot determine the decision class,continue the classification at the next level until the sequential process is terminated.The decision algorithm of the multi-category classification is given,and the calculation process of the algorithm is illustrated by an example.Then,the time overhead of classifying using this model non-incremental approach is large.In order to improve the efficiency of decision-making result,the updated equivalence class is calculated by using the calculation method of the object update equivalence class.By dynamically updating the equivalence class,the incremental method of the model is proposed to classify.Finally,the effectiveness and efficiency of the incremental method are verified by simulation experiments.