Experimental and machine learning method for evaluation of radiation risk from a research nuclear reactor

This study assessed the impact of the 10 MW Research Reactor on radiation levels in Islamabad, focusing on potential risks to nuclear facilities. Eight years of data (2013–2020) were collected using Thermoluminescent Dosimeters (TLD-200) installed near the reactor. Gamma spectrometry was used to measure soil radioactivity. The mean ambient dose rate was 0.21 ± 0.02 μSv h −1 , higher than previously reported values of 0.08 ± 0.01 μSv h −1 and 0.16 ± 0.03 μSv h −1 . The average TLD dose rate (0.21 ± 0.02 μSv h −1 ) significantly exceeded the soil dose rate (0.03 μSv h −1 ). Risk Assessment calculation were done to evaluate the risk from long operation of research reactor. A gated recurrent unit (GRU) neural network was used to compare radiation levels near the reactor with background levels, aiming to streamline future assessments and reduce experimental costs. To enhance predictive analysis and provide a cost-effective tool for long-term monitoring, a Gated Recurrent Unit (GRU) neural network was developed and rigorously benchmarked against other models, including LSTM, XGBoost, and ARIMA. The GRU model demonstrated superior performance, achieving an R 2 score of 95.4 % on training data and 93.7 % on test data, with low error rates (MAE of 10.043 and 11.113; RMSE of 30.001 and 30.125, respectively). The model effectively captured long-term trends, forecasting rising radiological levels at specific sites, though it was less sensitive to short-term fluctuations. Its robustness was further confirmed by average percentage errors of ∼1.43 % for the trained dataset and less than 1 % for the test dataset. Insights into geophysical and meteorological influences on radionuclide distribution were also provided, offering valuable data for monitoring and management. Despite increased radiation levels, the study concluded no significant environmental or health risks from the reactor after 54 years of operation. • Ambient dose 0.21±0.02 μSv/h, higher than prior reports, likely from increased molybdenum production. • Safety confirmed: hazard indices indicate no significant health or environmental risk after 54 years. • GRU model accurately predicts long-term radiation trends, enabling cost-effective monitoring and planning. • Geographic and seasonal variability observed, yet all dose levels remain within established safe limits. • Innovation : Combines experimental data and machine learning for safe reactor management.