Fuzzy Decision Tree Algorithm In The Research And Application Of Rainfall Forecast

In the21st century, the country’s economic construction, social progress and people’s life or work are closely related to weather forecasting, especially be important for forecasting severe weather timely and accurate. At the same time, there has accumulated vast amounts of meteorological data as the rapid advances in the development of computer science and technology and Meteorological observation technology. It has become an important issue need to be addressed now according to excavate inherent law efficiently and accurate meteorological Institute to these massive, diverse, multidimensional, complex, continuous meteorological data.Classification algorithm based on decision tree is an important classification of data mining technology field. It can express mining structure be very easy to use the way of graphical attribute structure. Using the decision tree inductive learning to generate rules, has become the most common and effective method in the knowledge acquisition, and be an effective way to build an expert system. Despite there has a lot of ways to structure relevant decision tree with the in-depth study of the decision tree, the traditional and clearly decision tree induction learning did not meet the needs of acquisition automatically of uncertain knowledge in a system. In order to acquire related knowledge set automatically in this kind of uncertainty environment. The research on fuzzy decision tree inductive learning has become a current research focus.This thesis focuses on the field of meteorology and regards the real meteorological data of a southern local region as the research object, Having a decision classification of rainfall in some areas. The main work of this thesis includes the following aspects:1) Having a pre-processing for a certain area of the local meteorological data, and remove redundancy and weak correlation properties by calculating rainfall and other correlation of related properties, then prepare for subsequent model building work.2) For the pre-treatmented meteorological data, dviding the rainfall into different levels according to the rainfall that the National Weather Service shows how divided. The clustering algorithm of data mining is used in the fuzzy treatment of continuous attributes, thus reducing errors to the final result in discrete rigid division and building the appropriate membership functions by clustering the results, then having a fuzzy processing for continuous properties.3) Using the mutual information and conditional mutual information, considering the impact of conditions in selected properties to the candidate in the fuzzy division and select candidate properties as the test attribute, then construct a fuzzy decision tree.4) In this paper, constructing decision trees for meteorological interim data on the basis of fuzzy decision tree model, calculating the similarities of decision tree model and weighted by voting, integrating decision-making models to achieve incremental learning process.