Optimizing Heavy Metal Detection with Deep Neural Networks and Gaussian Mixture Clustering in Voltammetric Data

The accurate and efficient detection of heavy metal is of paramount importance, given their deleterious effects on human health and the environment. While traditional laboratory methods are highly accurate, they are not without challenges, including high cost, non-portability, and time-consuming procedures. This study introduces a novel framework that integrates deep learning and clustering techniques to enhance the detection and classification of heavy metal from voltammetric data. The data were collected using cyclic voltammetry (CV) methods with a glassy carbon electrode (GCE) and subsequently analyzed using five classifiers. The following classifiers were employed: Support Vector Machine (SVM), Random Forest, K-Nearest Neighbors (KNN), XGBoost, and Deep Neural Networks (DNN). A dataset comprising 750 samples, categorized into three groups (cadmium, non-heavy metal, and lead), was subjected to advanced preprocessing techniques, including data normalization and dimensionality reduction through principal component analysis (PCA). The DNN model demonstrated superior accuracy, achieving 99.87%. This highlights its capacity to process noisy and complex data more effectively than traditional methods. Additionally, Gaussian Mixture Models (GMM) were employed to cluster and identify anomalies within the data. Despite its high accuracy, this study discusses limitations such as computational complexity and potential challenges in highly contaminated environments. These results underscore the significance of developing portable, efficient detection systems for on-site applications, offering a notable advancement over traditional techniques.