An interpretable machine learning strategy for growth rate of compressive strength in superabsorbent polymer internally cured concrete

Abstract

Regarding address the complex strength development and poor interpretability of prediction models for superabsorbent polymer (SAP) in internally cured concrete, this study integrates experimental data from coconut shell cellulose/attapulgite composite-modified SAP concrete with literature data to construct a multi-feature database. An interpretable descriptor for the Growth Rate of Compressive Strength (GRCS) was developed using symbolic regression combined with traditional machine learning. The results show that Extra Water (EW) is the dominant factor contributing to strength loss. A concise and generalizable mapping formula for GRCS was derived. The study elucidates characteristic thresholds within the system and reveals interaction effects between different features. Through material design and process control, the negative influences of SAP and EW can be effectively mitigated or even eliminated. This work provides a quantifiable basis for optimizing the mechanical performance of SAP internally cured concrete under complex conditions.