The transition to low-carbon construction materials is pivotal for reducing the environmental footprint of the cement industry. One of the key challenges in this process is the reliable prediction of mechanical performance, particularly compressive strength, of novel cementitious mixtures incorporating alternative binders and supplementary cementitious materials. This study presents a machine learning-based framework for predicting the compressive strength of low-carbon cement mixtures over multiple curing times (1, 2, 7, and 28 days), leveraging a compositional dataset of experimental formulations. We explore and compare several machine learning techniques, including ensemble learning, support vector regression, and neural networks, to model the complex, nonlinear relationships between input features (such as component ratios, mixture design parameters, and curing age) and compressive strength outcomes. Special attention is given to feature selection, model interpretability, and cross-validation strategies to ensure robustness and generalizability. Our results demonstrate that data-driven models can achieve high predictive accuracy, enabling virtual screening of formulations and guiding experimental work towards compositions with optimal strength development and reduced carbon intensity. Furthermore, we discuss the integration of domain knowledge in model development to enhance extrapolation capability and promote material innovation. This work highlights the potential of artificial intelligence to accelerate the design and deployment of sustainable building materials and contributes to the broader goal of decarbonizing the construction sector through data-centric research in materials science and engineering.