Classification of SERS spectra for agrochemical detection using a neural network with engineered features
Mateo Frausto-Avila, Muhammad Ismail Elias, Jose Pablo Manriquez-Amavizca, María del Carmen González-López, Gonzalo Ramírez‐García, Mario A. Quiroz‐Juárez
Abstract
Abstract Surface-Enhanced Raman Spectroscopy (SERS) substrates offer a promising solution for the sensitive and specific detection of agrochemicals, enabling timely interventions to mitigate their harmful effects on humans and ecosystems. However, the analysis of SERS spectra can be challenging due to the complexity of interpreting the data, often requiring advanced computational tools and expertise. This limitation highlights the need for continued innovation in both SERS technology and data analysis methods to fully realize its potential in real-world applications. In this context, we present a machine-learning model based on a feedforward neural network for the rapid and accurate classification of SERS spectra. Our approach consists of a highly compact neural network combined with feature engineering that was trained using SERS patterns from experimental measurements. The spectra used to train this model were acquired using substrates made of gold nanostars, which were deposited onto aluminum foil via drop-casting. The model utilizes a compact two-layer architecture, with five Leaky ReLU neurons in the hidden layer and four softmax neurons in the output layer. This design ensures computational efficiency by using only dense layers for matrix-vector multiplications. Notably, we performed feature engineering to optimize the input data; specifically, we derived 20 key features from transformation functions applied to the SERS spectra. The model demonstrates strong predictive performance, achieving high precision and recall values across all classes, with an overall classification accuracy of 98.5% for organophosphate pesticides and their mixtures. Compared to other machine-learning algorithms, our approach offers reduced computational complexity while maintaining or exceeding the accuracy of more complex models. This makes the proposed model particularly suitable for deployment in resource-limited environments, providing an efficient and effective tool for agrochemical compound classification in diverse environmental and food matrices.