A strength prediction model using artificial intelligence for recycling waste tailings as cemented paste backfill

The recycling of waste tailings as cemented paste backfill (CPB) has attracted worldwide attention because of the increasing environmental awareness during mineral resources excavation. However, lots of mechanical tests are required to understand the strength development of CPB and its prediction under the combined effect of influencing variables is almost an unexplored field. This study proposes a strength prediction model integrating boosted regression trees (BRT) and particle swarm optimization (PSO), where the BRT algorithm was used for modelling the non-linear relationship between inputs and outputs and PSO was used for the BRT hyper-parameters tuning. An extensive mechanical experiment was performed to provide the dataset for the PSO-BRT model. This dataset contained unconfined compressive strength (UCS) results of 585 CPB specimens produced with a different combination of influencing variables, including the physical and chemical characteristics of tailings, the cement-tailings ratio, the solids content, and the curing time. 10-fold cross validation was used as the validation method, and performance measures were chosen as the mean squared error and the correlation coefficient. The results show that PSO was efficient in the hyper-parameters tuning of the BRT. The optimum BRT model was very accurate at predicting CPB strength. The relative importance of influencing variables was investigated, in which the cement-tailings ratio was found to be the most significant variable for CPB strength. This research indicates that more efficient reuse of waste tailings as CPB can be achieved by reducing the required number of mechanical experiments during engineering applications.