FGHA-Net: Hybrid Attention Learning for Froth Flotation Recovery Prediction in Mineral Beneficiation

Froth flotation recovery prediction plays an important role in mineral processing because it can improve operational efficiency, reduce mineral loss, and support more effective decision-making during flotation operations. However, accurate recovery prediction remains challenging because flotation processes are highly dynamic and nonlinear, sensor data may contain noise, and process variables often exhibit complex temporal dependencies. Although several machine learning and deep learning approaches have been proposed for flotation prediction, many existing models remain limited in their capabilities of capturing short-term and long-range relationships in sequential flotation process data. To address these challenges, this study proposes a Fine-Grained Hybrid Attention Network (FGHA-Net) for froth flotation recovery prediction. The flotation process data are first preprocessed to improve data quality and suitability for model training. Feature engineering techniques, including process trend extraction, moving average features, and recovery-rate difference analysis, are then applied to enhance the representation of process dynamics. The proposed FGHA-Net consists of a bidirectional long short-term memory (BiLSTM) branch and a Transformer branch for hybrid temporal feature learning. The BiLSTM branch captures short-term sequential dependencies, whereas the Transformer branch models long-range temporal relationships using multi-head self-attention. A Fine-Grained Attention Fusion module is further introduced to support cross-branch feature fusion, temporal attention weighting, and refinement of important process variables. Finally, the fused features are passed to a regression head to generate the predicted recovery value. The proposed model is trained using the Adam optimizer and evaluated using standard regression metrics. Experimental results show that FGHA-Net achieved strong predictive performance on the testing set, with an RMSE of 0.011484 and an R² score of 0.844612. These results indicate that the proposed model can effectively learn complex temporal dependency patterns from froth flotation process data and has potential for intelligent mineral-processing monitoring and decision-support applications.

@article{Thanoon2026,
  title     = {FGHA-Net: Hybrid Attention Learning for Froth Flotation Recovery Prediction in Mineral Beneficiation},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {7},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Mohammed I Thanoon and Fawaz Alanazi and Shaaban M. Shaaban and Asma A. Alhashmi and Abed Saif Ahmed Alghawli and Ahlem Fatnassi and Majed Nawaz and Abdulbasit A. Darem},
  doi       = {10.14569/IJACSA.2026.0170736},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170736}
}