Comparing the performance of DTID3 and DTID3-Smote methods in predicting the rain events with unbalanced classes

Prediction of rainfall events in a region is important for many aspects of life. However, the majority of datasets that predict rainfall events have an unbalanced distribution of observations in their classes, including the Prabumulih city dataset, South Sumatra. DTID3 provides very satisfactory performance in many cases of prediction, while the Smote technique is useful for balancing the distribution of data classes. This study aims to compare the performance of the DTID3 and DTID3-Smote methods in predicting rainfall events in Prabumulih City. The main contribution of this study compared to previous studies is that the DTID3 and Smote methods are used together to predict rainfall events, especially in Prabumulih City. Using training data from 2017-2022 and test data from 2023, the results show that the DTID3-Smote method has a better performance measure than the decision tree method in predicting rainfall events in Prabumulih City. In the decision tree method, the accuracy, precision, recall, specificity, and f1-score metrics are 73.56%, 81.91%, 50.94%, 91.22%, and 62.81%, respectively. In the decision tree-SMOTE method, the values ​​are respectively 74.66%, 82.61%, 53.44%, 91.22%, and 64.9%.