Obtaining water quality criteria (WQC) that align with environmental conditions is essential for protecting aquatic life. However, complexity arises from numerous species across taxonomic groups and various environmental factors in water systems. Machine learning (ML) technology provides potential solutions to address these issues. This study employed artificial neural networks (ANNs) to analyze the applicability of a comprehensive mixed species model influenced by ten water chemistry factors. The Sobol method evaluated sensitivity of these factors to copper toxicity. Results showed the mixed model achieved high predictive accuracy after encoding taxonomic information. Based on sensitivity analysis, a simplified model considering temperature, DOC, DIC, and EC/hardness was proposed considering the commonly used monitoring indicators in practice. Comparisons of WQC results among mixed models based on all factors, simplified factors, and original samples demonstrated that focusing on key influencing factors does not lead to significant fluctuations in WQC outcomes. Finally, ecological risks in China's Yellow and Yangtze Rivers were assessed. Risks in the Yellow River exceeded those in the Yangtze River. For both rivers, dry season risks were substantially higher than rainy season, reflecting seasonal copper content differences and emphasizing environmental factors' significance in WQC derivation. The study provides insights for future WQC research.