Machine Learning-Based Compressive Strength Prediction in Pervious Concrete

The construction industry significantly contributes to global sustainability challenges, producing 30–40 percent of global carbon dioxide emissions and consuming large amounts of natural resources. Pervious concrete has emerged as a sustainable alternative to conventional pavements due to its ability to promote stormwater infiltration and groundwater recharge. However, the absence of fine aggregates creates a highly porous structure that results in reduced compressive strength, limiting its broader structural use. Determining compressive strength traditionally requires destructive laboratory testing of concrete specimens, which demands considerable material, energy, and curing time, often up to 28 days—before results can be obtained. This makes iterative mix design and optimization both slow and resource intensive. To address this practical limitation, this study applies Machine Learning (ML) as a rapid, preliminary estimation tool capable of providing early predictions of compressive strength based on mix composition and curing parameters. Rather than replacing laboratory testing, the developed ML models serve as supportive decision-making tools, enabling engineers to assess potential strength outcomes before casting and curing physical specimens. This can reduce the number of trial batches produced, lower material consumption, and minimize the environmental footprint associated with repeated destructive testing. Multiple ML algorithms were trained and evaluated using data from existing literature and validated through laboratory testing. The results indicate that ML can provide reliable preliminary strength estimates, offering a faster and more resource-efficient approach to guiding mix design adjustments. By reducing the reliance on repeated 28-day test cycles, the integration of ML into previous concrete research supports more sustainable, cost-effective, and time-efficient material development practices.