Agro-food waste valorization for sustainable biosurfactant production: Integrating fed-batch fermentation with machine learning

Biosurfactants are promising eco-friendly alternatives to synthetic surfactants; however, their large-scale application is constrained by high production costs. A novel biosurfactant-producing bacterium, Pseudomonas sp. BSP1 was isolated from oil-contaminated soil and characterized as a glycolipid-producing strain. To enhance sustainability, peanut residue (PR) and pineapple peel hydrolysate (PPH) were used as agro-food waste substrates in fed-batch fermentation, with distributed feeding improving biosurfactant production compared to pulse feeding. In the distributed strategy, 3% PR was supplied initially, followed by 1% PPH per day for 3 days, whereas pulse feeding received 3% PPH as a single addition on day 1. Distributed feeding increased biosurfactant production from 2.021 ± 0.019 to 2.174 ± 0.035 g L −1 and EI 24 % from 56.85 ± 0.18 to 60.44 ± 0.51%. A modified Latin Hypercube Sampling (LHS) strategy combined with Extreme Gradient Boosting (XGBoost) was employed to optimize six process variables using 100 experimental runs. Gain-based feature importance and SHAP analyses indicated that pH contributed most to model predictions. The optimized conditions yielded a predicted biosurfactant production of 2.903 g L −1 and an emulsification index (EI 24 %) of 63.53%. Experimental validation closely matched the predictions, achieving 2.948 ± 0.071 g L −1 and 63.24 ± 0.99%, corresponding to a ∼35.6% increase in production compared to pre-optimized conditions. Time-course analysis identified day 6 as the appropriate harvesting point, with biosurfactant production and EI 24 % continuing to increase after biomass accumulation peaked on day 5. These findings demonstrate the potential of integrating agro-food waste valorization with machine learning-assisted bioprocess optimization for sustainable biosurfactant production.