Font Size: a A A

Research For Updating Knowledge In Hybrid Dynamical Information Systems

Posted on:2016-04-05Degree:DoctorType:Dissertation
Country:ChinaCandidate:A P CengFull Text:PDF
GTID:1108330485483271Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
With the development of the internet and the Internet of Things (IoT), more and more massive and hybrid data are collected in real applications. The data volume is expanding drastically with the time and then forms dynamical, hybrid and uncertain information systems. The core competi-tiveness of enterprises have gradually shifted to the knowledge mining from the HIS. However, it is very difficult to finding important knowledge timely and effectively. Therefore, knowledge dis-covery in dynamical and hybrid information systems has become a hot topic in machine learning and data mining.Now, rough set theory and fuzzy set theory have become two important mathematical tools in knowledge discovery. Rough set theory considers knowledge as the division of data with each divi-sion as a concept. The main idea of rough set theory is to describe fuzzy or uncertain concept using priori knowledge in knowledge base. Fuzzy set theory is a method that describes fuzziness with a membership function. It has a robustness that fuzzy set theory is applied to solve the real problems. The superiority of rough set theory lies in the fact that it doesn’t need any extra preparatory infor-mation while fuzzy set theory needs it. However rough set theory has its limitations that it can’t describe inaccurate or uncertain problems fully and effectively. The fusion of the two theories may play their respective advantages and more powerful models will be obtained, namely, rough fuzzy set model and fuzzy rough set model. Based on these two models and combining with the incre-mental technology, some knowledge updating problems in the hybrid and dynamical information systems are discussed in this thesis.The main research works are as follows:First, incremental maintenance of approximations under the variation of object set in categor-ical and fuzzy decision information systems are discussed. The changing mechanisms of equiva-lence relation are discussed firstly when samples are added or deleted. And then, the propositions of updating the approximations and their memberships of the rough fuzzy set are analyzed. Based on this, the corresponding incremental algorithms are constructed.Second, two hybrid distance formulas are designed to deal with the hybrid data (e.g., categor-ical, real-valued, boolean-valued, set-valued, interval-valued, unknown-valued, text, image, video, audio and sensor signal data) in hybrid information systems. Combining gaussian kernel function with the hybrid distance, a method of fuzzy information granulation is designed, and then a new gaussian kernel fuzzy rough set model is developed. The changing principles of fuzzy equivalence relation with the changing of the object set are developed. Based on the new gaussian kernel fuzzy rough set model, the algorithms of incremental updating approximations are presented.Third, the changing principles of fuzzy equivalence relation and the fuzzy granulation with the changing of the attribute set are discussed. Based on the gaussian kernel fuzzy rough set model, the approaches of incremental updating approximations are proposed. And then, some principles and algorithms of incremental feature selection are presented based on forward greedy search.At last, an incremental approach for updating approximations of fuzzy rough set under the variation of attribute values in hybrid information systems are presented. The changing mecha-nisms of equivalence relation are discussed firstly when conditional attribute values are altered. Some propositions and an algorithm for incremental updating approximations are proposed under the variation of conditional attribute values. And then, the principles of concept hierarchical gran-ulation, and the relations between the coarsening or refining of decision attribute values and the concept hierarchical granulation are discussed. Based on this, the changing mechanisms of the decision classes partitions in fuzzy rough set are analyzed, and some propositions and incremental updating algorithms of the approximations in fuzzy rough set are presented.Some experiments on several data sets from UCI are employed to verify the effectiveness of the above incremental algorithms and the superiority with respect to the non-incremental algorithm-s, respectively.
Keywords/Search Tags:Hybrid Information System, Hybrid Distance, Rough Fuzzy set, Fuzzy Rough set, Knowledge Discovery, Incremental Updating
PDF Full Text Request
Related items