<p>Concrete compressive strength is critical for understanding the material’s load-bearing capacity. Traditional estimation methods often rely solely on concrete mixture properties and typically lack the integration of domain-specific knowledge that can improve accuracy and reliability. Concrete porosity, the volume of voids within the concrete matrix, is a key factor influencing compressive strength, as it directly affects the material’s density. Higher porosity generally leads to reduced compressive strength, as the voids within the concrete matrix weaken its load-bearing capacity. Despite its importance, existing machine learning models in the literature do not incorporate porosity as an additional input for estimating the compressive strength of concrete. Measuring porosity often requires experimental work, which drops the benefits of the machine learning model, resulting in an infeasible solution to estimate the compressive strength of concrete. Accordingly, this study proposes a novel porosity-informed hybrid neural network model that integrates the mixture’s concrete mixture properties and porosity to estimate its compressive strength. This study aims to develop a two-step hybrid neural network model where the first step predicts the porosity of concrete, and the second step uses the predicted porosity alongside other concrete mixture properties to estimate the compressive strength. Another critical novelty of this study is adopting a self-normalizing neural network, which has not yet been fully adopted in this field even though it holds significant potential for improving estimation accuracy, given its benefits in training stability and optimization. This approach enhances the modeling accuracy by further embedding material science insights into the prediction process while being feasible and reliable. The study results showed a 35% improvement in the root mean square error over the traditional neural network model that does not integrate porosity as an input.</p>