Performance evaluation of supervised machine learning models for Pb2+ ions detection in water using AlGaN/GaN high electron mobility transistors

In this work, a comparative machine-learning-based framework is developed for the binary classification of Pb 2+ ions presence in aqueous medium using 2, 5-dimercapto 1,3,5, thiadiazol functionalized AlGaN/GaN high electron mobility transistor (HEMT)-based sensor response data. Here, three supervised learning algorithms: logistic regression (LR), support vector machine (SVM), and random forest (RF) are implemented and tested on an experimental dataset comprising time-domain current response features. Prior to model training, data preprocessing and class balancing are performed to mitigate dataset imbalance and improve classification reliability. Model performance is quantitatively evaluated using accuracy, precision, recall, F1-score, confusion matrices, and receiver operating characteristic (ROC) curve analysis with corresponding area under the curve (AUC) values. The results demonstrate that SVM and RF models achieve near-perfect classification performance with high robustness, significantly outperforming the LR model. The present work highlights the effectiveness of advanced supervised learning techniques for reliable Pb 2+ ions detection and provides insights into model selection for machine-learning-assisted Pb 2+ ions sensing using AlGaN/GaN HEMT devices for water quality monitoring applications.