A novel hybrid machine learning approach for biorefinery products in pesticide-rich wastewater

Microalgae can produce products with high economic value within the scope of the biorefinery concept. In this study, machine learning (ML) approaches were used to enhance the production of carotenoids and biodiesel from Chlorella minutissima cultured in pesticide-contaminated wastewater, including malathion (Mal), chlorpyrifos (Chl), cypermethrin (Cyp), and atrazine (Atr). The highest carotenoid content reached 8.73 mg/g biomass under specific pesticide stress and cultivation conditions, while biodiesel production attained a maximum value of 139 % under a distinct parameter combination. The hybrid model exhibited strong predictive performance (R²:0.89–0.95), effectively reproducing the experimental responses for both carotenoid (Y₁, mg/g) and biodiesel (Y₂, %) outputs. Model interpretation using SHAP analysis indicated that Mal was the dominant factor influencing carotenoid accumulation, whereas biodiesel production was governed by a more intricate interaction involving Mal, Chl, and LI. Compared to conventional single-output modeling approaches, the proposed hybrid framework enables the simultaneous identification of high-yield operating regions for multiple products under multi-pesticide stress. These data driven observations are consistent with established stress-response mechanisms in microalgae and demonstrate the capacity of ML-based modeling to support informed decision-making in multi-product biorefinery systems operating under complex wastewater conditions. The proposed modeling approach therefore offers practical insight into balancing wastewater remediation with the sustainable production of high-value microalgal bioproducts.