Applying machine learning model for predicting unconfined compressive strength of cemented paste backfill on scarce data

Abstract

The compressive strength of Cemented Paste Backfill CPB (Cemented Paste Backfill CPB) is an important mechanical property in evaluating the applicability of this mixture in the reinforcement of mine pits. This paper presents the use of two simple machine learning models to predict the compressive strength of CPB. Therefore, two machine learning algorithms including Gradient Boosting GB and Support Vector Regression SVR are used to predict the compressive strength of CPB. For building the machine learning model, 92 experimental data were collected from international publications. The dataset includes six input variables such as cement/tailing content C/T, solids content (%), specific gravity Gs, sieve hole size with cumulative 10% D10 (μm), coefficient of uniformity Cu, coefficient of curvature Cc. The superior performance of GB machine learning model over SVR machine learning model is verified by 200 times of Monte Carlo random simulation. Feature importance analysis shows the necessity of the inputs to enhance the performance of the GB model can be arranged in descending order as follows: C/T ratio of cement/tailing content > coefficient of uniformity Cu > solid content > sieve hole size with cumulative 10% D10 > coefficient of curvature Cc > specific gravity Gs.

Keywords: Machine Learning (ML); unconfined compressive strength; cement/tailing; cemented Paste Backfill (CPB).