Modification to Optimise the Use of Agricultural Waste (MSA and SCBA) as Partial Cement Replacements Using Reactivity Indices for Concrete Strength Prediction

The cement industry contributes significantly to global carbon emissions, necessitating a transition towards a net‐zero carbon footprint for the industry. This study examines the feasibility of combining two types of agricultural waste, maize straw ash (MSA) and sugarcane bagasse ash (SCBA), as a sustainable partial cement replacement in concrete production. The research addresses dual environmental benefits by reducing cement‐related CO 2 emissions (8% of global CO 2 ) while preventing the uncontrolled burning of agricultural waste, which contributes to air pollution and surface water contamination. The primary challenge addressed is the variation in chemical compositions of agricultural wastes, which has traditionally hindered the combined use of dissimilar waste materials in predictive modelling despite their pozzolanic properties. This research examines concrete containing supplementary cementitious materials at replacement levels of 0%, 5%, 10% and 15% under ambient conditions by utilising a comprehensive dataset of 663 samples from the literature (267 samples) and experimental fieldwork (396 samples). Two predictive modelling approaches were evaluated: oxide composition‐based parameters (incorporating reactivity modulus, silica modulus, lime modulus, alumina modulus and hydraulic modulus) versus constituent composition parameters (conventional concrete mix components). Four machine learning algorithms were assessed: artificial neural network (ANN), Gaussian process regression (GPR), ensemble methods, and support vector machine (SVM), with ANN having 70%, 15% and 15% for training, validation and testing, respectively. Other models were trained and tested with 90% and 10%, respectively. The model’s performance was evaluated using the coefficient of determination ( R 2 R ), root mean square error (RMSE) and mean absolute error (MAE). Results demonstrate superior performance of oxide composition‐based modelling, with ANN and GPR achieving optimal accuracy (RMSE = 1.05 MPa and 1.21 MPa, respectively) compared to constituent composition‐based modelling (best RMSE = 2.54 MPa with ANN). The oxide composition approach successfully harmonises multiple dissimilar supplementary materials in a single predictive model, overcoming limitations of constituent‐based approaches that are restricted to single materials. These findings provide practical tools for construction professionals to reliably predict concrete performance when incorporating pozzolanic agricultural waste replacements, potentially accelerating the adoption of sustainable alternatives in the construction industry while maintaining structural integrity and performance standards.