Influence of design factors on pore size distribution in pervious concrete using machine learning classification

Pervious concrete is a sustainable, green material that reduces stormwater runoff by allowing water to pass through its porous structure. Pore size distribution governs the porosity, permeability, mechanical performance, and clogging resistance of pervious concrete. This study examined how the aggregate-to-cement ratio (3.0–5.0) and compaction effort (0–60 blows) influence the pore size distribution of pervious concrete. Here, Image analysis techniques were used to assess both surface-connected and internal pores. MATLAB's Classification Learner models were employed to classify the effects of these design factors on the pore size distribution. Increasing compaction reduces pore size variability, resulting in a more uniform distribution and improved compressive strength, with an optimal pore size range of 10–14 mm2. The Bilayered Neural Network achieved the highest accuracy (96.7%) for classifying low compaction, whereas the accuracy decreased in the high compaction range, with the Neural Network (Wide) achieving 78.9%. SVM with a Fine Gaussian kernel was most effective in classifying A/C ratios, reaching 80% accuracy. Feature selection identified the logistic curve constants b and k as key differentiators for compaction and A/C ratios. The findings highlight the potential of machine learning in optimizing pervious concrete mixtures by predicting how design factors affect pore structure.