This study applies machine learning techniques to forest fire damage area. The study area was Sokcho, Gangwon-do, which occurred forest fire damage on April 4, 2019. We used Sentinel-2 images for detecting forest fire area. We adjusted four cases for train model based on the dNBR severity level classes. We used 4 SVM kernels, because the accuracy may vary depending on the structure of the data to learning. The training results showed that in all four cases, the SVM RBF model showed the highest accuracy. The SVM RBF model with the highest training accuracy was used in the test area classification process. During the verification process, we created 300 GTP using KOMPSAT-3 for verification. Verification results show that the test results using the SVM RBF model classified (82.67%) was more affected than dNBR classification result (80.67%). The results of the study suggested a direction to incorporate mechanical learning in the analysis process for detecting and classifying forest fire damage areas.