INTERPRETABLE CARBON NANO SENSOR MODELING FOR WATER POLLUTION CLASSIFICATION

Abstract

Electrochemical carbon nano sensors are promising for rapid water pollution monitoring, but translating multivariate sensor responses into reliable pollution level decisions remains challenging. This study develops an interpretable machine learning framework for electrochemical carbon nano sensor response modeling in water pollution classification. A public environmental nanosensor dataset containing 6,786 observations was used, with graphene, multi walled carbon nanotube based, and hybrid carbon nanomaterial sensor records. The target variable consisted of two pollution severity classes, namely Low and Medium. Random Forest, Extreme Gradient Boosting, Support Vector Machine, and Multilayer Perceptron were evaluated using accuracy, precision, recall, F1 score, confusion matrix, receiver operating characteristic analysis, and precision recall analysis. The Multilayer Perceptron achieved the highest accuracy of 98.38 percent and macro F1 score of 96.62 percent, while Extreme Gradient Boosting achieved 97.86 percent accuracy and was selected for deployment. Interpretability analysis identified lead concentration, mercury concentration, nitrogen dioxide concentration, benzene concentration, rolling mean, and rolling standard deviation as dominant predictors. The proposed framework supports explainable water pollution classification and provides a deployable prototype for intelligent environmental monitoring.