Porosity plays a significant role in determining the durability of concrete. This study uses five machine-learning techniques to forecast the porosity of high-performance concrete blended with admixtures. The dataset comprises 240 records, encompassing various identical concrete mix designs. The input variables include admixture dose, fly ash content, slag content, superplasticizer dosage, and aggregate content. Extreme learning machines (ELMs) demonstrated superior predictive accuracy among the evaluated K-nearest neighbors (KNNs), artificial neural networks (ANNs), decision trees (DT), and gradient boosting (GB) methods. The ELM model achieved the highest performance by incorporating regularization techniques to mitigate overfitting issues. Compared to traditional numerical correlation models for porosity prediction, the data-driven approach addresses challenges related to estimating the time-dependent hydration degree and delivers improved prediction accuracy.