Data Mining in Weather Prediction

• Main Authors: Ku Chien-Hua, 古建華
• Other Authors: 顏秀珍
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/08312389095375240640

碩士 === 輔仁大學 === 資訊工程學系 === 94 === Abstract With the development of the computer technology and the trend of data digitalizing, the concept of the database management makes it possible for the storage media to change from the early paper sheet to the recent electronic storage devices, which increases the conveniences in managing and storing data. Besides, the computer technology makes it possible to handle enormous amount of data in short time with its fast computing capability. It is such capability that makes more and more people put effort on the study to improve data processing and analyzing. Weather forecasting would be one of the best examples for such change. Years ago, people made prediction with long sheet which had recorded weather data on it, and the process has been improved in terms of making different types of devices and computer systems doing all the measures and data storage. The weather men can easily and quickly make prediction on the weather for next hour through simulating and analyzing the data stored. However, here comes the problem. How if the weather men are not so experienced in making analysis, would they still predict the weather condition accurately? Thus, how to create some simple rules to assist the weather men doing their analysis has become a new field where draw more and more people’s attention. More than predicting the weather condition next hour, for those who work on weather forecasting, the data on weather can be further applied to the techniques on datamining to dig out information that may be helpful for forecasting. The main goal of this paper is try to research on the weather date with various dataminig techniques and to find whether there are certain rules which are helpful to weather predicting. We applied the algorithms of neural network, C5, and C&RT on the weather data collected between the year of 1996 and 2002 to see if there were any rules for the raining day and sunny day respectively. The result turned out that there existed certain characteristic for it to be a raining day or a sunny day. When the visibility decreased and the overall amount of cloud increased, it got pretty good chance to rain. On the other hand, it could possibly be a sunny day when the visibility increased and the overall amount of cloud decreased. When we further applied such rules to predict the weather condition of the year 2003, and it was up to 90% matched.