Machine learning application for radon release prediction from the copper ore mining in Sin Quyen, Lao Cai, North Vietnam

The radon release prediction from radioactive-bearing mines during mineral processing and mining is an essential target. A simple one-hidden-layer artificial neural network (ANN) model was designed with low computation cost to train, reference and get optimum effectiveness in comparison with two-hidden-layer ANN, random forest and support vector machine models which was applied for Sin Quyen copper deposit. The result showed with values of MAPE = 1.12(%), RMSE = 2.79(Bq/m³), MABE = 2.10(%), R² = 0.990, r = 0.99, for training part; MAPE = 1.12(%), RMSE = 2.79(Bq/m³), MABE = 2.09(%), R² = 0.995, r = 0.997 for testing part. The gamma dose and distance were significantly more effective variables for the radon prediction than direction, coordinate, and uranium concentration factors. (author)