<![CDATA[Poverty is a socio-economic issue that remains a concern in Indonesia, with differences in development characteristics between districts/cities causing wide variations in indicators that are intercorrelated. Feature redundancy and the existence of extreme values have the potential to reduce the generalization ability of classification models and reduce the interpretability of results. Therefore, an approach is needed that not only produces high accuracy but is also capable of identifying the most relevant indicators. Therefore, an approach is needed that not only produces high accuracy but is also capable of identifying the most relevant indicators. This study aims to evaluate the effect of Recursive Feature Elimination (RFE) on the performance of Support Vector Machine (SVM) and Random Forest in classifying the poverty status of districts/cities in Indonesia. The dataset used consists of 514 observations with two target classes, namely non-poor and poor. The preprocessing stage included data cleaning and outlier handling using the IQR capping method, then the data was divided into 80% training data and 20% test data. Testing was conducted on four scenarios: SVM, SVM+RFE, Random Forest, and Random Forest+RFE. Evaluation used a confusion matrix, accuracy, precision, recall, and F1-score. The results show that RFE does not change the accuracy of SVM (0.971), but improves the performance of Random Forest from 0.981 to 0.99 and improves the precision of the minority class. The Random Forest+RFE combination is the most effective and efficient configuration for regional poverty classification.]]>
