Integrating external stressors in supervised machine learning algorithm achieves high accuracy to predict multi-species biological integrity index of aquaculture wastewater

Abstract

Monitoring and predicting the environmental impact of wastewater is essential for sustainable aquaculture. The environmental DNA metabarcoding-integrated supervised machine learning (SML) algorithm is an alternative method for ecological quality assessment and prediction. However, the ecological integrity of aquaculture wastewater and available effective input features for prediction remain unclear. Here, we used the multispecies biological integrity index (Ms-IBI) to provide a detailed categorization, identifying over half of the samples (53.85 %) as highly impacted, emphasizing the urgency to address ecological degradation; Ms-IBI emerged as a reliable label for SML models. By condensing 410 effective indicators and integrating 25 core operational taxonomic unit features with external stressors, an accuracy of 0.78 and R2 of 0.96 was achieved. Utilizing only external stressors yielded a comparably good performance with fewer input features, obtaining an accuracy of 0.74 and an R2 of 0.91. The integration of external stressors in this study highlights a practical predictive method that meets the ecological quality requirements of aquaculture wastewater, aiding the reversal of global ecological decline.