Predicting harvesting efficiency of microalgae with magnetic nanoparticles using machine learning models

The emerging magnetic flocculation (or harvesting) with magnetic nanoparticles (MNPs) is a promising technology for microalgae dewatering. However, unearthing MNPs with high harvesting efficiency (HE) toward diverse microalgae relies on laborious experiments so far, a robust approach therefore is urgently needed to preevaluate the harvesting power of MNPs. Here, we predicted HE using machine learning algorithms across 1151 data points, in which the properties of MNPs and microalgae, and conditions of magnetic flocculation were comprehensively considered. Among 8 machine learning algorithms, the optimal XGBoost model showcased the best predictive performance with a high coefficient of determination (0.932), a low mean square error (6.96 %), and a low mean absolute error (4.17 %) on the test dataset. The model was also verified by batch experiments, demonstrating its ability to estimate HE accurately. Further, the Shapley additive explanations approach was used to decipher how the model made predictions from local and global perspectives, and these interpretations may offer guidelines for both the rational design of MNPS and the selection of microalgae species in magnetic flocculation. This work highlights the introduction of machine learning models to predict the harvesting ability of diverse MNPs toward microalgae, paving the way for the utilization of microalgal biomass.