Case Dependent Feature Selection using Mean Decrease Accuracy for Convective Storm Identification
Hansoo Lee, Jonggeun Kim, Seunghwan Jung, Minseok Kim, Sungshin Kim · 2019
Weather forecasting is one of the most critical information that closely related to real-life because of its influences in human society, especially high-impact weather. Advanced observation devices allow increase forecasting accuracy, but they also remain problems to solve: analyze data precise and fast, and reflect knowledge of experts. Recently, machine learning-based approaches have been presenting solutions for those problems. On the other hands, the approaches need to consider several conditions, such as model and feature selection. In this paper, we focused on the feature selection for the convective storm identification by using the random forest and the mean decrease accuracy method. By derived variable importance index, it is possible to determine correlation coefficient related features are more significant than others.