The Application Of Data Mining In Kunming Meteorological Data Analysis

Over the past few decades, the meteorological department has accumulated a large number of meteorological data, but how to use this data effectively face to such a big data is a challenge for all the workers in the field. Obviously, it is hardly to deal with such a big data with artificial methods, but we can dig out some valuable information when use some techniques about data mining with the help of computer.Firstly, this paper have pretreated to the data, including the analysis and conducts of outliers and missing values, meanwhile, merging the data by months and trying to make an exploratory analysis with average monthly data, hope to find the variation law of main index. Then have used the CART decision tree algorithm to predict and validate the data that monthly mean temperature is continuous, and also used CART algorithm and C5.0algorithm to predict and make a comparative analysis with box separation with equal width and use K-means clustering to discrete, and discovered that the latter’s accuracy rate is higher which on the training set is97.49%and on the validation set is91.67%.Secondly, it is still used two neural network(MLP neural network and RBF neural network) model to predict the monthly mean temperature which is discrete, and found the MLP model is better than RBF, the accuracy rate on training set is98.47%, and on the validation set is up to100%, in addition, this paper have compared the MLP neural network model with decision tree on the basis of the continuous target variable, and found that the MLP model is better than CART decision tree in many aspects, such as absolute average error, minimum error, standard error, the square sum of error and so on. It is turned out that the neural network is feasible and effective on predicting continuous and discrete target variable, especially the MLP model has higher accurate rate about predicting temperature.