초록 열기/닫기 버튼
Low Level Wind Shear (LLWS) is a sudden change in wind speed and wind direction below 2,000 feet above the ground, which can cause economic and personal damages. In Korea, wind shear occurs most frequently at Jeju International Airport due to the favorable topographical influences, such as the coast and Mt. Hallasan. In this study, we deal with the severe LLWS binary classification based on the LLWS data occurred in winter 2016-18 near Jeju International Airport, derived from a Limited area ENsemble prediction System (LENS). Severe LLWS is a rare event and creates a data imbalance issue and class overlap, resulting in a high misclassification rate of classification models. To solve this problem, we used sampling techniques as follows: SMOTE (Synthetic Minority Oversampling Technique), ENN (Edited Nearest Neighbor), NCR (Neighborhood Cleaning Rule). In addition, we applied a hybrid technique that combines modified ENN and STMOE, and compared the performances by learning classification models. As a result, ENN, NCR and hybrid sampling provided improved classification performance compared to the raw ensembles in terms of recall and F1-score.
키워드열기/닫기 버튼
Low Level Wind Shear, binary classification, data imbalance, class overlap
