Toward transparent AI: Predicting strength of fly ash foam composite concrete using explainable ML models

Fly ash foam composite concrete (FFC) is a sustainable, lightweight alternative to traditional concrete. However, accurately predicting its compressive strength (CS) through conventional laboratory methods is challenging due to its non‐linear behavior induced by the addition of foam content and fly ash. Also, the laboratory determination of CS is time‐consuming and resource‐intensive. Since there are very few techniques available in the literature to precisely predict CS of FFC based on its mixture composition, this study aims to overcome these limitations by developing accurate prediction models for CS of FFC using machine learning techniques such as AdaBoost, gradient boosting regressor (GBR), and multi expression programming (MEP). The mixture composition data of FFC to be used for this purpose was collected from already published literature having inputs such as water‐to‐cement ratio, foam content, and concrete age. The machine learning models were trained and tested on the training and testing subsets of this data. The model performance was evaluated using error metrics, external validation, and k‐fold cross‐validation. Among the models, AdaBoost demonstrated the highest prediction accuracy with an R 2 value of 0.977, while MEP offered the unique advantage of producing an interpretable mathematical equation. Further analysis using SHAP, individual conditional expectation, and partial dependence plots, etc., highlighted that foam content, fly ash, and w/c are key factors to determine CS of FFC. To facilitate practical application, a user‐friendly graphical interface (GUI) was developed, enabling engineers and practitioners to estimate FFC's CS efficiently and accurately using the best‐performing model from this study. The findings from this study will help engineers to accurately determine CS of FFC, and the insights from explanatory analysis will aid in mixture composition optimization.