Research And Application Of A Heuristic Attribute Reduction Algorithm

Along with the great development of modern information technology and computer network, it was convenient for us to get access to all kinds of data, which had already congested our daily life and work. Should it be unpractical to process such numerous and complicated data, if the traditional manual mode was used. The question aforesaid had drawn our attention as how to fetch the useful information expeditiously and veraciously from massive data to help our decision-making and management. Undoubtedly, computer, the most speedy operation machine, could be the solution to process the information, which required more thorough and comprehensive research on the data base knowledge discovery where the redundant, default and neutral information could be a major obstacle.In 1982 Z.Pawlak, Poland mathematician, put forwards rough set theory. It could process vague and undefined data, of which the model was brief and intuitionistic without validateding information in advance. As the decision rules deduced from rough set theory was straightforward, it had been manipulated in business triumphantly. This paper focused on the critical issues of rough set theory during knowledge discovery, such as data preproeessing, reduction and rules reduction, among which reduction was specifically analyzed and concluded, and it was discovered that until now there was not an acknowledged and efficient reduction algorithm. Under this circumstance, the author brought forward ameliorated arithmetic based on discernibility matrix and the heuristic algorithm for attribute reduction, in order to reduce time complexity, improve arithmetic efficiency and obtain the best reduction.In the first place, the paper studied the development and resent research achievement of rough sets at home and abroad, and the research significance and theoretics of rough sets, and it analysed the rough set means which were applied to every phases of knowledge discovery. In the next place, it analysed several methods of discretization of numeric attribute in data preprocessing phase, with the emphasis attached to the Naive Scaler algorithm for the discretization of continuous attributes. About decision table reduction, especially the attribute reduction and the value reduction were also explored. In the terms of attributes reduction, the author analyzed the familiar algorithms of the attributes reduction based on rough set, and pointed out the existent problem. Based on this research, an improved algorithm was put forwards according to the discernibility matrix and heuristic algorithm for attributes reduction, which resulted in the reduction of space time complexity, and obtaining the best reduction. With respect to attribute value reduction, an improved algorithm was advanced grounded on the heuristic algorithm for value reduction, actualizing efficient rules-obtaining. Last but not least, By comparative analysis of UCI database, the aforementioned improved algorithms were of higher efficiency and obtain better reduction results.And the paper applied the improved reduction algorithms to the analysis of students' examination results, making scientific evaluation of the gained rules by which the potential factor affecting the students' performance was discovered, hence, learning suggestion could be made. The efficiency and feasibility of improved algorithms were therefore validated by practical application.