Machine learning-based prediction of carbohydrate productivity in continuous cultivation of Chlorella vulgaris

Predicting carbohydrate productivity in continuous microalgae cultivation systems remains a significant technical challenge due to the non-linear nature of metabolic pathways under multiple stresses. This study applied Machine Learning (ML) models to predict biomass and carbohydrate productivity in Chlorella vulgaris grown in continuous culture, utilizing 145 independents experimental sets. Linear (Multiple Linear Regression, Ridge, and LASSO) and non-linear (Random Forest, Artificial Neural Networks, and Support Vector Regression) models were evaluated, integrating nutritional (N and P), environmental (light intensity and optical density), and operational (residence time) variables. Model optimization was carried out via grid search with 5-fold cross-validation and an 80/20 data split to ensure robustness and prevent overfitting. Results showed that non-linear models significantly outperformed traditional methods. Random Forest emerged as the most effective algorithm, achieving an R² of 0.9072 and RMSE of 0.0518 for biomass productivity, and an R² of 0.9304 and RMSE of 0.0187 for carbohydrate productivity. These findings demonstrate the potential of ML as a "virtual sensor" for real-time control and optimization of large-scale industrial bioprocesses, enabling immediate operational adjustments without reliance on time-consuming laboratory analyses.