Research On Sequential Three-way Decision With Autonomous Error Correction And Its Optimization Model

In the era of big data,data is growing explosively.Therefore,how to conduct knowledge discovery effectively from the massive,dynamic,high-dimensional,and sparse data is a difficult problem faced by artificial intelligence.From data to knowledge and ultimately to decision provides a new paradigm for people to deal with complex problems,in which the problem of uncertainty is one of the hot spots.As one model of granular computing,sequential three-way decisions provide people with a progressive decision-making process for dealing with uncertain problems.It has received considerable attention due to its flexible processing mechanism and efficient decision-making.By continuously acquiring new information,the uncertain problems can be solved gradually in the process of transforming from coarse granularity layer to fine granularity layer.However,with the in-depth development of the model,some issues worthy of study need to be further explored.For example,in the existing sequential three-way decision model,only the boundary region is used as the universe of next layer for further subdivision without positive or negative regions processed.In addition,the decision thresholds obtained based on the artificial cost parameters have a certain subjectivity.Therefore,designing a more reasonable and adaptive threshold structure is an urgent problem to be solved.In response to the above two issues,this thesis has done the following work:(1)For the first problem,two types of misclassification phenomena exist objectively in the classification process.Therefore,from the perspective of granules subdivision,this thesis analyzes and discusses the change rule of the error classification rate of positive and negative regions with equivalence classes proportional and disproportionate subdivision.On this basis,two types of effective error correction and two types of effective classification are defined.In addition,since the objects in the boundary region may have a high possibility of misclassification,hence,based on the k-means clustering technology,a new sequential three-way decision model with autonomous error correction is proposed by selecting a portion of the equivalence classes in the positive and negative regions near the two sides of the boundary region.Experimental results show that the proposed model has smaller error classification rate compared with traditional sequential three-way decisions.(2)For the second problem,this thesis formulates a game between the error classification rate of the positive and negative regions and the uncertainty of the boundary region by introducing a game theory in the sequential three-way decision model,so that a pair of adaptive decision thresholds can be obtained from each granularity layer.However,since the error classification rate and uncertainty are two discrete random variables with different dimensions and criterions,to eliminate the influence of dimensions and criterions to calculate the total incomes,a comprehensive payoff function normalized by the L-2 norm was given.Moreover,under the strategy profile of two-player game,the game stopping conditions are presented based on pure strategy Nash equilibrium analysis.Finally,a series of comparative experiments were performed to compare the effect of the proposed model with multi-objective decision method,and the experimental results verify the validity and rationality of the proposed model's decision thresholds.