Research On Continuous Attributes Discretization And Rules Extracting Basesd Rough Set

As a new soft computing method, rough set theory connects the knowledge with sorts and supports a mathematics tool which accords with the human's cognizing for the classification problem of inadequacy data. The knowledge discovery technology based on rough set has become the hot topic of the decision science. With the development of the social economy and the improvement of the people's living, the weather problem has been one of the focuses which have been attended by the people. How to get the veracious calamity weather rules by the observed data has been attended more and more by the weather people. We have researched several key problems about rough set, and applied the results to the weather problem, and gotten better results.First, we give a GA measure of discretization which is proposed first here. It can supply a gap of row-column measure which gets a result affected by the position of the attributes and breakpoints. Furthermore, we have compared several kinds of discretization algorithms and applied them to the weather data. The experimental results have proved that the GA measure is a relative effective measure, and the combination of adding-minusing classes measure and row-column measure is better.Second, we research two attributes reduction measures. One is based on the discriminated matrics, the other is based on GA. We use many new ways to make the algorithms more effective, such as reducing dimension and sparsiting elements of the discriminated matrics, effectively selecting elements of the positive examples set and the counter examples set for the jointing-spreading matrics, using a new selecting operator, and so on. The experiments have proved that they reduce the algorithmic time and space complexity greatly, and improve the algorithmic efficiency.Third, we propose a new way of decision rules extracting based on the discriminated matrics. It can use the reduction's middle and last results effectively, so reduce the algorithmic time and space complexity effectively. The validity has been proved by the experimental results we obtained.At last, We develop a software platform of data discretization and rules extracting based on rough set. It can be used for the pretreatment of data mining. We analyse the weather data by the system and get satisfying results.