Research On Prediction And Decision-Making Methods In Multi-Scale Information Systems From Three-Way Decision Perspective

Prediction and decision-making,as the priority research of intelligent sciences in the context of big data,play a significant role in promoting the development and progress of society.Prediction,as a premise and basis for decision making,is essential for scientific decision making.At present,the research on prediction and decision-making focuses on how to effectively mine complex and uncertain big data information,and enhances the rationality and scientificity of the prediction and decision-making process,so as to provide technical support for realistic prediction and decision-making problems.As a granular computation and knowledge discovery methodology to effectively measure complexity and uncertainty,three-way decision can effectively measure the uncertainty and complexity in the decision-making process by introducing a delayed strategy.Therefore,this paper will conduct an investigation of the prediction and decision-making methods for multi-scale information systems from three-way decision perspective.In particular,the main research of this article contains two aspects as follows:On one hand,a multivariate prediction method with three-way decision is conducted in a multi-scale information system.Specifically,this paper addresses the weaknesses and deficiencies of existing prediction researches in dealing with prediction problems in the context of big data,and develops a prediction framework that can autonomously mine the relationship between features and models,which not only effectively improves the prediction performance and generalization ability of models,but also increases the scientificity and rationality of the prediction process and results.Firstly,to exploit the multi-scale prediction information adequately,scale rules are utilized to achieve information fusion.Secondly,a novel score function is constructed for each feature with regret theory to measure the importance of each feature,which sufficiently takes into account the human irrational behavior in the prediction process.Furthermore,to eliminate dimensional disaster and reduce the prediction cost,an adaptive sequential three-way feature selection model is developed for the selection of feature subsets.Subsequently,this paper considers that the importance of different samples for the prediction system tends to be different,and an improved extreme learning machine model is constructed in combination.On the basis of the above work,a multivariate prediction system that can simultaneously perform feature subset selection and model parameter optimization is developed.Finally,the prediction system is applied to an actual multi-scale rate-estimation problem,and the experimental results indicate that the developed prediction system has better performance than several existing prediction methods.On the other hand,a multi-attribute decision-making problem research with three-way decision is conducted in a multi-scale information system.Considering that available multi-attribute decision-making methods are proposed under single-scale data information generally,which possesses great limitations for dealing with complex and uncertain decision problems in the context of big data.Moreover,human is the subject of decision-making process,and its irrational behavior influences the decision outcome to a certain extent.Accordingly,this paper constructs a novel multi-attribute decision-making method from a three-way decision perspective.Firstly,to mine effective decision information from multi-scale information systems,an optimal sub-system selection mechanism is constructed via matrix theory in this paper.Secondly,a quantification of fuzzy information under optimal scale combination is constructed based on interval number,which characterizes nicely the ambiguity and uncertainty of information in the era of big data.Then,a measure of conditional probabilities and thresholds based on regret theory is constructed from the perspective of three-way decision,which not only provides a new idea for obtaining conditional probabilities and thresholds,but also enhances the rationality and scientificity of the decision process.Finally,a three-way multi-attribute decision-making methodology with classification and ranking is established.The experimental results on real numerical cases illustrate that the proposed method is feasible and effective,and more scientific and universal compared with the traditional multi-attribute decision method.In summary,this paper constructs a research on prediction and decision-making methods for multi-scale information systems from the perspective of three-way decision,which not only enriches the research results of prediction and decision-making theories,but also provides new ideas for solving complex prediction and decision-making problems in the context of big data.