Data-Driven Mercury Detection Using Neural Networks and Nanocluster-Based Sensors

Mercury contamination in drinking water remains a serious threat to environmental ecosystems and public health, driving the need for sensitive and reliable sensing systems. Metallic nanoclusters have gained attention for such applications due to their distinctive physicochemical properties, including high surface reactivity and nanoscale dimensions. However, selecting the most suitable nanocluster-based sensor from previously reported candidates is challenging, as sensing performance depends on several interacting factors such as metal composition, ligand characteristics, and measurement conditions. In this study, we developed a data-driven machine learning framework to screen nanocluster candidates for mercury detection in drinking water and to model how physicochemical descriptors relate to sensing success. A dataset of ۱۰۸ samples was compiled, where each candidate was characterized using descriptors including hydrogen-bonding ability, sulfur-containing functionality, nitrogen and oxygen content, acid-base characteristics, electrical charge, the metal's tendency for amalgamation with mercury, and cluster size. The prediction task was formulated as a supervised binary classification problem. Three machine learning models Random Forest (RF), Linear Support Vector Machine (SVM), and Multi-Layer Perceptron (MLP) were assessed using stratified ۵-fold and ۱۰-fold cross-validation. Model performance was evaluated using accuracy, precision, recall, F۱-score, ROC-AUC, and confusion matrices. Particular emphasis was placed on minimizing false negatives, since overlooking a truly effective sensing nanoparticle is undesirable in environmental monitoring. Among the evaluated models, the MLP consistently achieved the best overall performance, yielding the highest recall and the fewest false negatives across both validation settings. The superior performance of the MLP, together with the comparatively weaker results of the linear SVM, suggests that mercury-sensing performance is governed by nonlinear interactions among multiple nanoparticle descriptors rather than simple linear relationships. These findings demonstrate that AI-assisted screening can improve the identification of high-performance nanocluster-based sensors and provide insight into the physicochemical factors associated with successful mercury detection.

کلیدواژه ها:

Neural Networks ، Machine learning ، Nanocluster-based sensing ، Mercury detection