Hierarchical discrimination of pyrethroid types and varieties using a ceria nanozyme array with machine learning.

The toxicity and environmental impact of pyrethroid pesticides (PYRs) are closely related to their chemical structures and are particularly influenced by the presence or absence of a cyano group (Type I/Type II). Current rapid detection methods are unable to achieve hierarchical identification of PYRs, from broad class discrimination down to single-compound identification. In this work, by screening nanozymes with different ligand types, four cerium-based nanozymes bearing aromatic carboxylic acid ligands that exhibited significant responses to PYRs were selected, and a four-channel colorimetric array was constructed based on their differential responses to these PYRs. Subsequently, a multi-output random forest (Mul-RF) was introduced, enabling hierarchical recognition from Type I/II (cyano group absence/presence) down to eight specific varieties. Independent of PYR concentration, both category-level and variety-level identification yielded excellent results. After completing category identification, a Mul-RF model was further utilized to reliably predict of PYR concentrations. In spike tests with spinach, soil, lake water, and cotton hulls, the array correctly achieved hierarchical identification from class to individual compound, with no cross-category misclassification or variety confusion. This hierarchical pesticide identification strategy based on a nanozyme array and machine learning provides a new paradigm for precise risk assessment and on-site stepwise screening of PYRs in environmental and food samples.