A machine learning model to predict unconfined compressive strength of alkali-activated slag-based cemented paste backfill

The unconfined compressive strength (UCS) of alkali-activated slag (AAS)-based cemented paste backfill (CPB) is influenced by multiple design parameters. However, the experimental methods are limited to understanding the relationships between a single design parameter and the UCS, independently of each other. Although machine learning (ML) methods have proven efficient in understanding relationships between multiple parameters and the UCS of ordinary Portland cement (OPC)-based CPB, there is a lack of ML research on AAS-based CPB. In this study, two ensemble ML methods, comprising gradient boosting regression (GBR) and random forest (RF), were built on a dataset collected from literature alongside two other single ML methods, support vector regression (SVR) and artificial neural network (ANN). The results revealed that the ensemble learning methods outperformed the single learning methods in predicting the UCS of AAS-based CPB. Relative importance analysis based on the best-performing model (GBR) indicated that curing time and water-to-binder ratio were the most critical input parameters in the model. Finally, the GBR model with the highest accuracy was proposed for the UCS predictions of AAS-based CPB.