Machine Assisted Optimization-Driven Nano Thin-Film Platform for Quad-Metal Heavy Metal Detection and Quantification

Abstract

Trace detection of heavy metals in complex environmental water matrices remains challenging due to ultra-low concentrations, matrix interference, and the need for rapid and selective analysis. This study presents an optimization-driven surface-enhanced Raman
spectroscopy (SERS) platform integrated with machine learning for ultra-sensitive detection of Pb²⁺, Cd²⁺, Hg²⁺, and As³⁺ across ultrapure, synthetic freshwater, tap, and river water matrices. A cysteine-functionalized AgNR substrate enabled detection down to 1 ppt and a wide dynamic range up to 1 ppm, with stable, reproducible Raman signals and minimal intensity decay (<6%) over 14 days. To enhance quantification and overcome matrix effects, SVR and SVM models were employed, achieving strong predictive performance with R² of 0.981–0.993, MAE <5%, F1-scores of ~0.95–0.99, and classification accuracy exceeding 95% across all metals. Variability analysis further confirmed reproducibility with signal fluctuations below ±6%, demonstrating a robust and reliable analytical framework for rapid and selective monitoring of trace heavy metals in diverse water systems.