Abstract: Current prediction models for dust concentration in open-pit mines usually rely on predefined indicators and parameters to estimate dust levels. However, the open-pit mining environment is complex and highly variable, and the key factors influencing dust concentration are numerous and difficult to clearly define. As a result, existing prediction models show limitations in prediction accuracy and generalization ability, and they often ignore the deep spatiotemporal features embedded in the data, making it difficult to comprehensively characterize the variation patterns of dust concentration. To address these issues, a dust concentration prediction model for open-pit mines based on a Long Short-Term Memory network (LSTM) - Convolutional Neural Network (CNN) was proposed. Pearson correlation analysis was used to select humidity, noise, stripping volume, and wind speed as input indicators for dust concentration prediction. After preprocessing, these indicators were fed into multiple parallel CNN units with an attention mechanism, which extracted local spatial features at different scales. The attention mechanism was used to weight the extracted features, enhancing feature representations that were more strongly related to dust concentration while suppressing redundant or noisy information. The attention-enhanced features were then reorganized into a time-series format, and the gated mechanism of the LSTM captured the temporal dynamic patterns and long-term dependencies of dust concentration. Finally, dust concentration was predicted through a fully connected layer. Experimental results showed that, compared with single models including LSTM, CNN, and Random Forest (RF), the proposed LSTM-CNN model improved the coefficient of determination R² by 7.85%, 12.91%, and 23.49%, respectively, reduced the Root Mean Square Error (RMSE) by 17.81%, 45.76%, and 33.35%, and reduced the Mean Absolute Error (MAE) by 26.48%, 25.56%, and 24.52%. Compared with hybrid models RF-SVR and RF-GRU, the LSTM-CNN model improved R² by 2.89% and 4.79%, reduced RMSE by 9.15% and 14.12%, and reduced MAE by 11.40% and 16.53%, respectively.