Machine-learning-optimized T3C2Tx/Bi2Se3 nanoflower-modified screen-printed electrodes for electrochemical detection of trace Pb2+ and Cd2+ in artificial sweat.

In the field of non-invasive detection methods, sweat analysis has been increasingly capturing the interest of researchers due to its potential for clinical and health monitoring applications. While the detection of various substances in sweat has been extensively documented, the detection of heavy metal ions such as cadmium (Cd2+) and lead (Pb2+) is challenging due to their low concentrations, often falling below 100 ppb. Sweat testing also faces challenges such as variability in sweat production and evaporation rates, biological fouling, and sensor sensitivity decay, which significantly affect the accurate measurement of biomarker levels in sweat. Traditionally, machine learning in the biosensing domain is often employed for final data analysis and fitting, but its application in optimizing experimental conditions is less common. In this study, we developed a sensing platform based on Ti3C2Tx/Bi2Se3 nanoflower-modified screen-printed carbon electrodes (SPCE) for the detection of Cd2+ and Pb2+ in simulated sweat. The sensor demonstrates a broad detection range and relatively low limits of detection (LOD), with a linear detection range of 10-150 ppb and an LOD of 2.88 ppb for Cd2+, and a linear detection range of 5-150 ppb and an LOD of 3.45 ppb for Pb2+. In addition, an LSBoost-NGO machine-learning framework was employed to model the dependence of the stripping current on key experimental variables and to identify operating conditions that maximize analytical performance. These results demonstrate the potential of Ti3C2Tx/Bi2Se3/SPCEs combined with machine-learning-guided optimization for sweat-based monitoring of trace heavy metal ion.