Metaheuristic Optimization to Improve Machine Learning in Raman Spectroscopic-based Detection of Foodborne Pathogens

Abstract

Accurate and reliable determination of foodborne pathogens (FBPs) is necessary for food safety. Spectroscopic methods such as FT-IR and Raman spectroscopy are among the label-free and sensitive methods for diagnosing FBPs. Although Raman spectroscopy equipped with confocal microscopy is developed for multiplex detection of FBPs, machine learning methods optimized by advanced optimization algorithms can be useful for the efficient determination of FBPs in food. In this study, genetic algorithm (GA) and particle swarm optimization (PSO) were used to optimize the architecture of artificial neural networks (ANNs) to predict the type of FBPs based on their Raman data. Raman spectra of single cells of 12 common strains from five genera were obtained to create a dataset. The results showed that the average accuracy of GA-ANN and PSO-ANN hybrid models was 0.89 and 0.93, respectively. Moreover, ATCC 14028 and ATCC 19112, the strains of Shigella and Listeria bacteria, were predicted with the highest performance (0.96) based on the Raman spectra of their corresponding cells. The method presented in this study included Raman spectroscopy combined with neuron-based machine learning methods for the FBP efficient diagnosis.