Integrating spectroscopy with machine learning and deep learning for monitoring mung plant responses to silicon dioxide nanoparticles

Abstract

This study investigates the potential of integration of confocal micro-Raman and UV–Vis spectroscopy with machine learning and deep learning algorithms to assess biochemical responses of mung bean plants exposed to silicon dioxide nanoparticles (SiO₂ NPs) at varying concentrations. The analysis of acquired Raman spectral data reveals a concentration dependent pattern where low concentrations (0.2–0.6 mM) reduce the intensities of key biomolecules such as carotenoids, lignin, pectin, protein, carbohydrate, and cellulose, while higher concentrations (1.2–1.4 mM) trigger enhancement in intensities. The estimation of photosynthetic pigments using UV–Vis spectroscopy complements the Raman spectroscopy results, with chlorophyll and carotenoid levels decreasing at lower concentrations before significantly increasing. Among computational approaches, the application of dimensionality reduction techniques such as LDA- significantly improve the performance of clustering algorithms learnings like AGNES (RI = 1.00), DBSCAN (RI = 0.99), and k-means (RI = 1.00) and deep learning models, achieving high classification accuracy. Supervised algorithms like random forest and support vector machine perform optimally without dimensionality reduction, showing accuracies of 78 % and 79 % respectively. This integrated spectroscopy-computational approach offers a non-invasive, label-free, and robust framework for monitoring plant-nanomaterial interactions.