Concrete compressive strength prediction is crucial in structural engineering, particularly for local and non-standard aggregate mixtures whose characteristics differ from those assumed by conventional design codes. This study maps the application of ML regression algorithms for concrete compressive strength prediction using non-standard materials, specifically Indonesian local aggregates, and IoT–ML integration. Literature was searched using the PRISMA 2020 protocol from various databases, with 119 core references synthesized in a matrix of source, non-standard material, best algorithm, and evaluation metrics. The synthesis results show three main findings. First, gradient boosting and ensemble algorithms tend to be superior, with XGBoost achieving an R² of up to 0.998, bagging regressor 0.97, and stacking reducing RMSE by 4.2%. Second, ANN is effective for waste materials and complex mixtures, with an R² of up to 0.9888 and MAPE of 2.87%. Third, IoT–ML integration enables real-time compressive strength prediction, with an error of 0.76–7.83% and an R² of up to 0.996. However, ML research on local Indonesian aggregates, such as laterite, coral rock, coral pumice, and sea sand, remains limited. This indicates a research gap in the need for a standardized, validated regression ML framework for local Indonesian aggregates and integrated with IoT architectures.
Copyrights © 2026