Adaptation And Convergence For Incremental Concept Drift

Concept drift is a research hotspot and difficulty research problem in data stream mining,and it is a common phenomenon in the real world and virtual world.With the rapid development of Internet technology,it has been attracted more and more attention from academia and the public.At present,concept drift doesn't have a unified and clear definition.It mainly refers to the unpredictable changes in a part of the connotation information or characteristic information of the information system,which makes the original classifier lower accuracy,and even leads to undesirable consequences such as misjudgment,commonly appears in recommendation systems,finance,and decision-making.Rough set theory is a mathematical tool for incomplete and uncertain data,it can effectively analyze and process all kinds of inaccurate and incomplete information in information systems and can be used without any prior knowledge.Under the circumstances,to analyze and reason the data can get a relatively objective and fair processing result or description.For its unique advantages,rough set is gradually applied to concept drift detection.At present,most researcher attitudes to concept drift is still at the level of simple avoidance.There are few studies on the conditions and trends of concept drift,and it is a blind spot for how to judge cognitive convergence.Rapid and subtle concept drift often occurs in the data stream.Concept drift has little impact on the decision-making system in a short time,so it is often ignored.However,some effects will be gradually superimposed,resulting in a decrease in the accuracy of the decision-making system.At present,there is a lack of suitable and effective methods to solve such problems.The research content of this thesis is: Firstly,it tries to combine the incremental learning with the traditional rough set model,and to combine the advantages of the two theories to build an incremental concept drift detection model that can adapt to data stream.It detects rapid and subtle concept drift in the data stream,and continuously improves the classification accuracy through iterative training.Theoretical analysis and experimental results show that the algorithm and research strategy proposed in this thesis is effective and feasible.The new model is more sensitive and efficient than other static models in detecting concept drift and learning new knowledge.The good effect of the new model is also the future of rough set theory.How to promote the application of big data analysis and data stream processing provides an effective and feasible new idea.Secondly,the new model proposed in this thesis is used as a research tool to study the concept drift from the perspective of the overall decision system.In the research results of previous scholars,the nature of cognitive convergence and the judgment standard was further improved and adjusted,and the influence and connection between concept drift and cognitive convergence are analyzed.From a cognitive point of view,the results of this article can explain why when different data sets or different parts of the same data set are used for classification,the reasons for the large difference in the before and after results can also explain the differences in opinions in real life.The innovation points of this paper are summarized as following:1.Based on the theory of incremental learning and rough set,an incremental concept drift detection and adaptation model is proposed.2.From the perspective of rough set,this thesis further improves and adjusts the properties and criteria of cognitive convergence in a decision system and analyzes the influence and connection between concept drift and cognitive convergence.3.The connotation and influence of concept drift are further extended.