Three Kinds Of Cost Under The Environment For Sensitive Attribute Selection

Cost-sensitive learning is a hot research spot in the data mining,and budget constraint satisfaction problem is one of the problems in artificial intelligence and machine learning field.In recent years,the attribute selection problem with the minimum test cost is a research focus for cost-sensitive learning.Moreover,in real applications,due to the resource we can afford for any practical problem is often limited.There is a budget constraint under certain resource conditions.Therefore,considering cost-sensitive attribute selection problem and budget constrain problem have been applied in many field and it is great significance.In addition,the current cost-sensitive algorithm generally used fixed static misclassification cost,which only meet the needs of the experiment and forward-looking,and cannot adapt to the same kind of sample size distribution dataset of classification model learning.Therefore,aiming at the shortcomings of the static misclassification cost,how to design the dynamic mechanism of misclassification cost is favored by more and more scholars.Based on the attribute selection problem of the minimum test cost,attribute selection problem of the budget constrain problem and the attribute selection problem of the dynamic misclassification cost are studied,they are main innovations are as follows.Firstly,we studied the cost-sensitive attribute selection problem for the minimum cost,and minimum cost simply consider the test cost.A logarithmic weighted algorithm is employed to find the attribute selection with the minimal test cost in this paper.The experimental results show that the effect of the new algorithm is superior to the existing algorithm in most cases.Secondly,we studied cost-sensitive attribute selection problem under the budget constrain.The budget constraint is that the test cost constraint which means the maximal test cost one can afford is less than the minimal test cost.To be more specific,under the condition of test cost constraint,the aim is to find an optimal feature subset which can keep the highestdegree of the information of decision systems.Under the optimal feature subset with test cost budget constraint,a simulated annealing algorithm is designed in this paper.Experimental results show that the algorithm for we design can obtain good effectiveness and efficiency of the experimental results,the experimental results better than the existing heuristic algorithm and genetic algorithm.Finally,we studied the cost-sensitive attribute selection problem for the dynamic misclassification cost.What is more,four optimal misclassification cost functions are designed,which incorporate the relationship between the minority class and majority class as well as the test cost,forming the objective representative misclassification cost of space.Automatically searches suitable misclassification costs for ever dataset,and it can better approximately real dataset of misclassification cost.