Whirlwind Classification with Imbalanced Upper Air Data Handling using SMOTE Algorithm and SVM Classifier

Abstract

Whirlwind is a natural disaster that often occurs and is difficult to predict from some time before. Early identification is needed to prevent a lot of casualties and losses. Whirlwind caused by instability in the atmosphere. Instability in the atmosphere usually occurs at the beginning of the day and the whirlwind can be identified based on the upper air parameter which can represent atmospheric instability. The purpose of this research is to optimize SVM classification with SMOTE algorithm to handling problems in imbalanced data and this research can minimize casualties and losses or also be a breakthrough for disaster-prone areas to be given early warning. The process for identifying whirlwind has several stages, namely pre-processing, imbalanced data handling, and classification. Pre-processing is normalized data. Whirlwind data has a classification problem, namely inter-class data that is not balanced so it needs to be corrected using the SMOTE Algorithm. Research on the identification of whirlwind using a combination of SMOTE-SVM produces the best accuracy is 98.8 %. When compared to data without the SMOTE algorithm the results obtained are better if the SMOTE method is applied. The specificity value is also better when given the SMOTE method. Based on these results it can be concluded that SMOTE can overcome the problem of imbalanced data in the upper air data by increasing the value of the classification of the whirlwind.