Research On Data Mining And Forecasting Of Meteorology Data

In the 21st century of the knowledge and economic time and facing with the fact of bursting data but poor knowledge, Data mining has been put forward and applied in many fields of database management. Data Mining is a deep-level data analysis method, including association analysis, clustering analysis, isolating analysis and classification and forecasting etc. Classification is a very important problem of Data Mining, which is the summary of constructing classification model and applying it to classify the important data category and forecast the tendency.Firstly, the thesis simply analyzes the present condition of data mining and classification algorithm. Then some improved classification algorithm have been suggested, which emphasized on studying on and improving data pretreatment, neural network and decision tree.It is a necessary part of data mining of data pretreatment that cleaning and inducing data and providing object data for classification algorithm. The thesis analyzes some data pretreatment method and proposes a new method with applying PLS (Partial Least Square) in PCA (Primary Component Analysis) in order to decrease data' s dimension. At the same time it has been used that the fuzzy clustering method in disperse important data category to form classification attribution.Neural network is a recently classification. BP neural network, RBF neural network and LVQ neural network have been used in constructing network classification in this thesis. This kind of classification is good in precise classify but poor in extracting rules. The thesis pays more attention on improving the precision of classification but less on extracting rules.Decision Tree is a usual data classification method. Several common decision tree algorithms have been analyzed and it has been found out that there is great relationship between generating decision tree and selecting data sample. So based on selecting sample a better decision tree algorithm has been suggested, whichwithout changing the traditional method' s mind but emphasizes on adjusting sample with analyzing input and output data sample in order to generate a superior decision tree rules.At the end, the thesis summarizes the former methods with its advantages and inadequacy, then it has been suggested the fowling study and research object.