Research On The Decision Fusion Method Of Incomplete Information System Based On Generalized Rough Set Theory

Decision-making is the core content of management science. With the rapid development of computer software and hardware, as well as the wide applications of database technology, humans need to acquire and deal with increasing data in practice, and the related decisionmakings become more and more complicated consequently. These new problems in the era of big data put forward higher requirements to the ability of intelligent information processing. Discovering potential and useful information from the vast, unordered and corrupted data to evaluate objects, make decisions and predict future goals becomes a hot area of research. Humans are constantly looking for some new data analysis methods to mine the application potential from information systems effectively. The evolutionary process from data to information, and then from information to decisions, makes up the concept of decision fusion. The research on the decision fusion method not only combines the characteristics and advantages of information fusion, data mining, and knowledge discovery in intelligent information processing, but also focuses on the target of “decision information extraction” from some complicated information systems, e.g. incomplete information system. The related theory and methodology can be studied and summarized systematically to propose some complete and feasible technical schemes finally.The research of this thesis is carried out from the perspectives of “decision” and “fusion”. Combined with the thought of information fusion and the algorithm of generalized rough set theory, the decision fusion method is intensively studied in the application backgrounds of system evaluation, assistant decision-making, failure analysis, and etc. The main points and contributions of the thesis are as follows.1) Research on the development of generalized rough set theory and the algorithm of attribute reduction.Firstly, some important operators and sets are proved to be consistent in both traditional rough set and the generalized one. Thus, the unified research framework of generalized rough set theory is established to creat the research foundation of the whole thesis. And then, from the viewpoint of informatics, the concept of “entropy” is introduced for the definitions of information entropy and condition entropy. By comparing the differences among various of information entropies and condition entropies, their properties in the information systems are proved respectively. Thus, the quantification of the uncertainty of information systems may get better by the applications of these entropies. By calculating the values of condition entropy describing different combinations of condition attributes relative to the same decision attributes, the E-condition entropy-based heuristic algorithm of attribute reduction is modified for generalized rough set and becomes the core algorithm of the decision fusion method in this thesis.2) Research on the model-based decision fusion method of incomplete information system.The incompleteness caused by data missing and uncertain addes difficulties to the decision rule extraction and restricts the process of high quality decision fusion. The incomplete information system, regarded as the main research object of the decision fusion method in this thesis, can be dealt with by two kinds of solutions in the unified research framework of generalized rough set theory. They are the model-based technique and the imputation-based technique. Here, the model-based technique is studied firstly. The performance of data classification models with the tolerance relation, the non-symmetry similarity relation, the limited tolerance relation, and the connection degree tolerance relation are carefully compared in the incomplete information system. Then, the direct model-based decision fusion method of incomplete information system is formed. Following that, according to the connection degree tolerance relation, an improved metric called α-classification quality of approximation is suggested to measure the quality of decision fusion. Thus, the influence of changes in the volume of missing data becomes assessable and a new assessment method for the impact of missing data to the quality of decision fusion is established.3) Research on the interval-valued information system and the information filling technique.The construction of interval-valued information system provides a favorable platform for the study of information filling technique. In the unified research framework of generalized rough set theory, the interval similarity degree is incorporated to the definition of information entropy, and a new generalized information entropy called H’-information entropy is defined in the thesis to build the relationship between the length of interval-valued data and the uncertainty of the system. Finally, the missing data is reasonably replaced by the estimated interval-valued data, and the incomplete information system is converted to the complete information system which can be dealt with by some traditional decision fusion methods. That is the incompleteness is solved indirectly by the information filling technique at this moment.4) Research on the improvement of data classification model and the robustness of decision fusion method.The improvement of data classification model is an important step to enhance the performance of decision fusion. After data preprocessing, the related statistics of real-valued data can be obtained according to the statistical theory. Then, the T-test method is incorporated to the data classification model to establish a new hypothesis testing-based model. Thus, the application scope, the calculation accuracy, and the noise adaptability of data classification model are clearly improved. The robustness of the decision fusion method based on generalized rough set theory is significantly enhanced for the big data as well. That is the new hypothesis testing-based data classification model may provide a better solution to the decision fusion of incomplete information system in the big data environment.5) Process design and software implementation of the decision fusion based on generalized rough set theory.The general process of proposed decision fusion method based on generalized rough set theory is also designed in the thesis, which includes the methods and purposes of each step from data acquisition to decision application in the decision fusion method. The main functions and process are implemented by setting up “the demonstration platform of decision fusion method based on generalized rough set theory”, and the transformation from theoretical results to practical applications has been promoted.The above research achievements will be presented in detail in the five central chapters of the thesis. Several complete application examples and the comparison with some other decision fusion methods are provided in each chapter to verify the research significance and the application value of the proposed generalized rough set theory-based decision fusion method.