Land fire is a complex and serious problem affecting several human life sectors. Stimulating the occurrence of land fires caused by long dry seasons and erratic rainy seasons causes plants to dry up and die. Illegal land burning by irresponsible people for expansion and other interests. This study uses quantitative methods. The variables used are vegetation index values obtained from Landsat 8 OLI images. The result of this research is that data classified using the random forest algorithm at RStudio produces a default number of 500 trees, and the number of variables tested in each split is 2. So, the estimated error rate of 2 out of the bag is 0.47%. Prediction and confusion matrix - train 403 data are less risky "0" (middle risk) of a fire, and there are 22 data of high risk of "1" (high risk) of a fire, with an accuracy value of 0.9914, 1. While the second prediction of the confusion matrix - test data, there are 167 data less risky "0" (middle risk) of fire. There is 1 data with a high risk of “1” (high risk) of a fire. There are seven high-risk data "1" (high risk) of fire, with an accuracy value of 0.9943, indicating a more accurate assessment of model accuracy, and the range is still good, around 96% to 99%. There are several incorrect classifications in the prediction and confusion matrix – train, which is slightly higher than the predictions of the two-confusion matrix – test, and sensitivity: 1.0000, which is the same in both prediction and confusion matrix – train & tes.