Machine learning assessment of mechanical properties of oil palm shell concrete

Abstract

The escalating global demand for natural coarse aggregates in concrete production has intensified concerns over resource depletion and environmental degradation. As a sustainable alternative, oil palm shell (OPS) has emerged as a promising lightweight aggregate, yet its mechanical performance requires robust predictive modeling to facilitate wider adoption. This study develops and evaluates advanced data-driven models to predict the uniaxial compressive strength (UCS) of OPS concrete. An extensive experimental database of 377 specimens was compiled, incorporating key mix design variables including cement, water, sand, OPS content, superplasticizer dosage, and curing age. Multiple artificial neural network (ANN) architectures were systematically compared against conventional multi-linear regression (MLR). Results demonstrate that the ANN outperformed MLR by a wide margin, with the optimal two-hidden-layer architecture (6−13−3−1) achieving superior predictive accuracy (MAPE = 6.983 %, R = 0.977). Sensitivity analysis further highlighted water, cement, and OPS content as the dominant parameters influencing UCS. Beyond establishing OPS as a viable eco-efficient aggregate, this work underscores the power of machine learning in optimizing mix design and accelerating the practical deployment of sustainable concretes.