Meta-analysis and interpretable machine learning model of organic removal and power generation in photosynthetic microbial fuel cells

Photosynthetic microbial fuel cells (PMFCs) offer a promising approach for simultaneous wastewater treatment and energy recovery, but the reported performances vary widely due to heterogeneous reactor designs and operating conditions. A meta-analysis of 136 observations from 44 studies revealed substantial dispersion in chemical oxygen demand (COD) removal and power generation, even under comparable operating conditions. Although monotonic associations were observed between influent COD and COD removal efficiency (Spearman ρ = 0.78) and between hydraulic retention time and power generation (ρ = −0.61), considerable performance variability remained within similar conditions, indicating that PMFC is governed by multivariate interactions rather than single-parameter effects. Multivariate machine learning modeling integrated these nonlinear dependencies and identified stable governing drivers. Influent COD concentration and hydraulic retention time emerged as dominant predictors for COD removal efficiency, whereas power generation normalized by COD removal was regulated by external resistance, with light intensity exerting a complementary influence. Ensemble models achieved robust predictive performance (average test R 2 of 0.95 for COD removal efficiency and 0.92 for normalized power generation), demonstrating cross-study generalizability despite fragmented reporting. These findings suggest that PMFC performance is fundamentally constrained by substrate loading and electrochemical operating conditions, thereby providing a quantitative basis for data-informed design across diverse system configurations.