Based On Rough Set Theory, Machine Learning

Machine learning is one of the most important fields of research in Artificial Intelligence. Machine learning is the unique path of machine acquiring knowledge and is a subject in which it is researched that how to use machine to simulate mankind's studying activities. It is the basic goal of machine learning that computers can have the ability to study, simulate and fulfill mankind's studying activities. Machine learning is a method by which we can study how computers identify existing knowledge, acquire new knowledge, improve its function and achieve its perfection. Machine learning has successful application in many domains. However, with the development of technology and improvement of demand, some problems have appeared such as knowledge acquisition and how to deal with fuzzy problems or uncertainty. At present, the level of machine learning cannot meet the needs of practice. Ultimate reason is that there is not breakthrough in the theory methods of system's knowledge express and management. The problem that machine learning must solve is that how to acquire inexact knowledge and relation, at the basis of which, a correct conclusion must be drawn.Rough sets theory is a new learning hotspot in the field of Artificial Intelligence. It can effectively deal with incomplete, uncertain knowledge express and reasoning. Moreover, its validity has been proved in many fields. Rough sets theory need not any prior knowledge or information, and can analyze and dispose imprecise, inconsistent, incomplete datum. It discloses potential disciples, pick up useful information and reduce information by finding hidden relation among data. Rough sets also provide knowledge reduction which is a tool to acquire knowledge from data automatically. Therefore, it has a wide prospect to introduce knowledge acquisition and knowledge reduction in machine learning.Attribute reduction and value reduction are the basis of learning from examples based on rough sets. This thesis studies the problem of attribute reduction and puts forward an improved reduction algorithm based on the dependence and importance of attribute in order to acquire brief decision rules. It makes learning from examples have high-efficiency. Besides, we discuss concepts and research methods of several important problems which rough sets theory refers to in machine learning. Because database has two dynamic changes, (1) adding examples (2) adding condition attribute without changing in examples. An incremental algorithm for attribute reduction based on rough sets is proposed. Moreover, the attribute reduction results obtained respectively from this algorithm and the traditional algorithm are identical. However, in comparison with the traditional algorithm, the time complexity of this algorithm is lower and its classification quality is better. Then, after making a deep analysis of the problems existing in the inductive learning algorithms based on discernibility matrix and traditional decision matrix, a new decision matrix and an algorithm based on this matrix for inductive learning and incremental learning in the environment of incremental data set are proposed. The main idea is to divide the decision system into several subsystems on the basis of the decision classes and construct the new decision matrix, and then to transform the inductive learning for the decision system into the incremental learning based on the new decision matrix. It not only could solve the problem of inductive learning on the incremental data sets, but also could considerably reduce the size of traditional decision matrix and avoid the repeated computation in traditional decision matrix algorithm. The example analysis and experimental results illustrate the rationality and validity of the new algorithm based on the strategy of dividing and conquering. Therefore, the two methods can adjust to the dynamic changes of database in machine learning and meet the demand of dynamic machine learning.