Experimental and nature-inspired modeling of lipid accumulation in microalgal biomass for biofuel production

Accurate determination of lipid contents (LC) in microalgae is important in optimizing biofuels production processes, but it is facing challenges because of the complex interactions of the biological and environmental factors. The current paper introduces an experimental study integrated with nature-inspired learning for modelling LC in Neochloris oleoabundans biomass during different nutritional and ecological settings to develop and produce biofuel systems that are sustainable using microalgae. The quality of the data was verified by linear and modified Akima cubic interpolation of the missing values, the Gaussian smoothing with the factor of 0.25, and removing the outliers via the Generalized Extreme Studentized Deviate (GESD) method. The spline interpolation was used to fill the observed outliers. All the variables were statistically tested and found to be reliable and stationary using the Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) tests guided by the Schwarz Criterion (SCI), Akaike Information Criterion (AIC) and Bartlett kernel Criterion (BK). Four models, including Extreme Gradient Boosting (XGBoost), Extreme Learning Machine (ELM), ELM optimized with Whale Optimization Algorithm (ELM-WOA), and CatBoost were developed based on three different feature sets which serve as the working hypothesis of this study. The outcomes showed that the CatBoost model with feature combination C3 has the best predictive power with a testing phase RMSE = 0.1593. The feature relevance was compared through Gain-based feature importance based on the internal mechanism of the XGBoost model, dependency analysis, analysis of variance (ANOVA), and Shapley Additive explanations (SHAP), which all determined the main predictors (cultivation time, biomass concentration, and ETR pH).