Frost Forecasting considering Geographical Characteristics

Regional accuracy was examined using extreme gradient boosting (XGBoost) to improve frost prediction accuracy, and differences by region were found. When the points divided into two groups with weather variables, Group 1 had a coastal climate high minimum temperature, humidity, wind speed 2 exhibited relatively inland characteristics. We calculated in found that precision recall scores areas (Group 1) significantly lower than those 2). Geographic elements (distance from nearest coast height) added as variables accuracy. In addition, considering continuity of occurrence, method reflecting occurrence previous day variable synthetic minority oversampling technique (SMOTE) pretreatment used increase learning ability.