Affiliations:
Department of Information Management, National Yunlin University of Science and Technology, Yunlin, Taiwan
Correspondence:
[*]
Corresponding author. Ching-Hsue Cheng, Department of Information Management, National Yunlin University of
Science and Technology, Yunlin, Taiwan. Tel.: +886 5 5342601/5312; Fax: +886 5 5312077; E-mail: [email protected].
Abstract: Rainfall forecasting has been a popular research topic. Precise rainfall prediction can not only assist water management in region of water scarcity, but it can also warn or alleviate the effects of excess or insufficient rainfall. As a result of the advancement in information technology, current prediction methods are more diverse and sophisticated; however they require significant amounts of resources, and time are costly, and the forecast outcomes are still very uncertain. Therefore, this study proposed a novel rainfall forecast model, which combined the proposed integrated non-linear attributes selection method with support vector regression (SVR) to enhance the forecast performance. First, the proposed integrated non-linear attribute selection method was employed to determine the important attributes that affect rainfall in the mountainous region of Taiwan, and then, the selected attribute data were input into the SVR model to train the rainfall forecast model. To assess the prediction performance of the proposed model, this study collected rainfall data from 2005 to 2014 at monitoring stations in the Taiwanese mountains, and compared the proposed model results with those of the listing models. Experimental results show that the proposed model outperforms the listing models in terms of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE).
Keywords: Component, feature selection, rainfall forecast, SVR, time series
