Design and development of an expert system based on random forest machine learning model for identifying reuse potential and predicting long-term utilization of Disused Sealed Radioactive Sources (DSRS)

Effective and efficient radioactive waste management is a major challenge for the Radioactive Waste Treatment Plant (WWTP). This research focuses on developing an expert system based on the Machine Learning (ML) Random Forest model to identify reuse potential and predict the long-term utilization goals of Unused Radioactive Substances (ZRTTD). The main objective of this research is to develop a model that is able to process radioactive waste data effectively, provide appropriate reuse recommendations, and predict changes in activities and utilization goals over the next 100 years. The data used includes original data and synthetic data generated using the Bayesian Network method. Data preprocessing involved normalization and handling class imbalance using the Synthetic Minority Over-sampling Technique (SMOTE). The Random Forest model was then trained and evaluated using accuracy, F1-Score, AUC- ROC, precision, and recall metrics for classification, and Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-Squared for regression. The evaluation results show that the Random Forest model performs very well. For classification, the average value of F1-Score is 0.90, recall is 0.88, precision is 0.98, and AUC-ROC is 0.99. For regression, the average MSE value is 2.673803e-09, RMSE is 5.2e-05, MAE is 5e-06, and R-Squared is 0.999999. Predictions of radionuclide activity decline are consistent with their respective half-lives, and the model is able to provide accurate predictions of long-term utilization goals. This research provides practical insights for radioactive waste management by IPLRs and can be an effective tool in supporting decisions related to radioactive waste management in Indonesia. (author)