South Korea has been in a truce with North Korea since 1953, and conventional artillery firing training are still needed considering its geopolitical location. Based on the weather condition, the decision is made whether to conduct artillery training, which might cause forest fires. Wildfires caused by military training not only devastate forests but also worsen public opinion on national defense. While forest fires triggered by artillery training cause substantial damage, the number of occurrences is very low, making it difficult to construct a prediction model in a general manner. In this paper, the weighted support vector machine for imbalanced data is applied to predict the risk of forest fire due to artillery firing training. We employ an over-sampling technique based on a probability distribution for imbalanced data and applied a weighted support vector machine algorithm that enforces a misclassification cost of the minority class. This study not only considerably reduces the probability of forest fires occurring during conventional artillery fire training in the Republic of Korea Army (ROKA) but also encourages the development of practical approaches for wildfires prediction in countries with climates similar to the Korean Peninsula. Furthermore, our proposed method can contribute to the study of the classification model for various imbalanced data. Through Monte Carlo simulations, we demonstrate that our proposed method achieves significantly higher accuracy than traditional methods.