Optimization of point-of-use ceramic water filters: linking pore morphology to filtration performance via a hybrid design-of-experiments–machine learning framework

The paper introduces a combined design of experiments, response surface methodology, and machine learning framework for the optimization of low-cost ceramic water filters manufactured from locally sourced clays. Three types of clays (black cotton, red, and alluvial) combined with sawdust (40–70 wt%) were evaluated over a firing temperature range of 850–950 °C. Multiresponse optimization identified the final optimal conditions at 66 wt% sawdust and a firing temperature of 950 °C, yielding an overall desirability index of 0.92. At this point, microbial removal exceeded 95%, and effective porosity was approximately 45%. Scanning electron microscopy confirmed that these performance peaks correlate with the formation of interconnected macropore networks (∼10–150 μm) and microcracks. While higher firing temperatures encouraged densification, the random forest model demonstrated superior predictive accuracy (R2 = 0.97) in capturing the nonlinear relationship between fabrication parameters and microbial rejection. This hybrid approach provides a scalable pathway for decentralized water treatment in line with UN Sustainable Development Goals 6 and 12.