Interpretable prediction method of Matte grade based on GA-BiLSTM and SHAP

Aiming at the problem that the ice-copper grade prediction process is difficult to effectively capture the long-term dependency relationship between data, which leads to low prediction accuracy, it is proposed to construct a model using the powerful long-term memory capability and bidirectional information prediction ability of bidirectional long and short-term memory network (BiLSTM) model, and further optimize the model by using the genetic algorithm (GA) to achieve accurate prediction of ice-copper grades. First, the genetic algorithm was used to find the optimal hyperparameter combination of the model. Second, the predicted values of the BiLSTM model optimized by the genetic algorithm are compared with the predicted values of the existing methods and the unoptimized BiLSTM model using different evaluation indexes. Finally, the SHAP method is used to give the overall and local interpretation of the prediction results of the BiLSTM model. The experimental results show that the RMSE of the BiLSTM model optimized by the genetic algorithm is reduced by at least $3.05 \%$, the MAE is reduced by at least $2.75 \%$, the MAPE is reduced by at least $3.29 \%$, and the $\mathrm{R}^{2}$ is increased by at least 1.51%, which verifies the effectiveness of the proposed method in solving the problem of ice-copper grade prediction, and the SHAP method is able to explain the output of the BiLSTM model efficiently and increase the transparency of the model. The SHAP method can effectively explain the output of the BiLSTM model and increase the transparency of the model.