In the era of artificial intelligence, energy is crucial, and nuclear power generation is efficient but also has potential dangers. Therefore, radiation levels around nuclear power plants should be closely monitored. The Nuclear Safety Commission has established only 63 radiation monitoring stations across Taiwan. Due to certain equipment malfunctions, the detected radiation values may be missing or abnormal. This study aims to use machine learning methods to estimate missing radiation values in the event of monitoring station malfunctions, thereby improving the continuity and usability of radiation data records. Effective feature selection is performed on the data, followed by data preprocessing to enhance the effectiveness of model training. The process involves model selection and construction, where the optimal architecture is identified and fine-tuned through hyperparameter adjustments. The performance of the proposed model is evaluated using the coefficient of determination R2 that can achieve 0.92 or above.