Performance Evaluation of ANN and Multi-linear Regression Models in Predicting Concrete Slump

Concrete slump is a critical measure of fresh concrete workability that influences placement and durability. This paper presents an artificial neural network (ANN) model to predict concrete slump based on mixture composition, and compares its performance to a traditional multiple linear regression model (MLR). A dataset of 732 concrete mix designs including varying cement, aggregate, water, admixture, and supplementary cementitious material contents was used to train different ANN models. The ANN captured the complex non-linear relationships between mix ingredients and slump, converging to a stable solution. Preliminary findings show that the optimal ANN achieved a validation and test losses of 0.072 and 0.09, a coefficient of determination R2 = 0.82 and an a20 index of 0.95 in predicting concrete slump demonstrating high fidelity to experimental values. Benchmarking against multiple linear regression (MLR) confirms a marked improvement in both predictive accuracy and robustness. The findings underscore the potential of machine learning techniques in concrete materials engineering, enabling more efficient mix design optimization and reducing the need for exhaustive trial batching.