Predicting rubberized concrete compressive strength using machine learning: A feature importance and partial dependence analysis

Abstract

Rubberized concrete is a material that is both ecologically friendly and sustainable, and it has been finding more and more usage in building applications recently. In this study, a machine learning model, namely LightGBM, is developed to predict the compressive strength (CS) of rubberized concrete using 11 input parameters. The performance of the model is measured using a number of different statistical criteria after it has been trained on a dataset containing 275 samples. In order to evaluate the impact that each input parameter has on the CS, feature importance and partial dependency plots (PDP) are used as analytical tools. According to the findings, the superplasticizer, chipped rubber, crumb rubber, coarse aggregate, fine aggregate, and water content all have a significant impact on the CS of rubberized concrete. On the other hand, the results indicate that the cement content, slag/fly ash content, and type of CS have a relatively minor effect. In addition to this, the PDP offers insights into the manner in which the input parameters have an effect on the CS of rubberized concrete. Overall, the developed model and analytic techniques may be used as a helpful tool in forecasting the CS of rubberized concrete and improving its mix design for a variety of construction applications.