Data Mining Solution for Detecting the Behaviour of Climate Changes

Abstract

Climate changing pattern detection is a critical problem because of its continues dynamic behaviour. Climate simulation models and advanced observation gathering techniques are introducing with the development of technology which produce thousands of output data by providing a good opportunity for data miners. But the data gathering speed exceeds the data analysing rate. This research is trying to discover that hidden knowledge out from a data source which generated as an output of weather research and forecasting model and investigate the applicability of such models to the Sri Lankan context. The main objectives of this research to identify the factors which contribute to climate changes by investigating climate models use by other countries. These points will be further discussed under literature ndings. Then by checking the relevance of these features to the Sri Lankan context, we are going to select relevant attribute subset from complete data source by using attribute selection methodologies. Then by performing appropriate data mining techniques we are proceeding the pattern identi cation part. The basic data mining techniques that we are going to use for this research is association rule mining and J48 tree classi cation technique. Association rule mining technique just produce the association rules related to non rainy periods. But J48 tree classi cation was able to generate rules related to all rainfall ranges with a good level of accuracy. According to the literature review, Neural network approach can be consider as the best weather and climate modelling technique. To test and evaluate J48 decision tree classi er we have used k fold cross validation technique and multilayer perceptron network. Results of this research shows that J48 tree classi cation technique is capable to produce better classi cation comparing to multilayer perceptron network, accuracy vice as well as performance wise. WRF(Weather Research and Forecast) model is capable to generate huge amount of weather related attributes as its model outputs. So this is good opportunity to weather researching community. In future we can come up with a weather prediction model by using the knowledge we obtained.