The interpretable machine learning strategy deepens the application of cemented tailings backfill material with tailings-cement-fly ash in engineering

The massive accumulation of mine tailings (MT) poses a serious environmental challenge in mining areas, and their utilization in cemented tailings backfill (CTB) offers an effective solution for solid waste reuse and safe underground backfilling. However, traditional methods for evaluating the unconfined compressive strength (UCS) of CTB composed of MT, ordinary Portland cement (OPC), and fly ash (FA) are time-consuming and limited in applicability, highlighting the need for an efficient and accurate predictive approach capable of capturing complex nonlinear interactions. In this study, a machine learning (ML) model was developed based on 702 multi-source samples, achieving excellent predictive performance on the testing set ( R 2 = 0.970) after systematic comparative selection and optimization. Feature importance analysis showed that curing time contributed the most to UCS prediction (32.181%), followed by OPC mass ratio (22.893%) and FA mass ratio (11.396%). SHapley Additive exPlanations analysis revealed a pronounced staged nonlinear effect of curing time and a clear threshold effect of water-to-binder ratio at approximately 18%. Validation on 24 independent samples confirmed the model’s engineering applicability ( R 2 = 0.880), and a user-friendly graphical user interface was developed to improve practicality. This study provides a robust ML framework for UCS prediction of complex cemented materials, offering strong technical support for scientific design and safety management of CTB systems in mining engineering.