Resampling Techniques in Rainfall Classification of Banjarbaru using Decision Tree Method

Abstract

Continuous heavy rains, such as in 2021, can cause flood emergencies in various areas of Banjarbaru. Therefore, classification modeling is needed to predict rainfall classes based on climate parameters. The problem faced in the classification case is the unbalanced class distribution. Class imbalance occurs when the minority class is much smaller than the majority class. This research aims to compare three resampling techniques, namely Random Undersampling, Random Oversampling, and SMOTE, in handling imbalanced rainfall data in Banjarbaru using the Decision Tree model. The comparison methods used were sensitivity, specificity, and G-Mean values. The research results show that the best model is the Decision tree model with the Random Undersampling technique because it provides the highest G-Mean value and sensitivity and specificity values above 70%. Based on this model, the variables that can separate the Rainy and Cloudy classes are Minimum temperature, Maximum temperature, and Sunshine duration, with the best separator being Maximum Temperature.