Abstract: This article challenges the stereotype that North Korea’s foreign policy is difficult to predict and thus can only be subjected to ex post study. Based on the naive Bayesian method, we establish a short‐term prediction model for North Korea’s nuclear and missile tests using international news reports from North Korean media between 2006 and 2018 as a dataset. The test results show that the overall accuracy rate of the model’s predictions of North Korean historical activity is greater than 80%, and its robustness is strong. To solve the problem of relatively delayed data collection, we use the Seasonal Autoregressive Integrated Moving Average (SARIMA) time series analysis method to simulate the values of feature sets. The estimated data are statistically reliable, and the prediction accuracy is high. This study proves that although the DPRK is extremely closed, it is possible to make relatively accurate predictions of its behavior using appropriate methods. The modeling approach in this paper can provide inspiration for developing general approaches to national behavior prediction.