Different supervised and unsupervised classification approaches based on visible/near infrared spectral analysis for discrimination of microbial contaminated lettuce samples: Case study on E. coli ATCC

Abstract

Escherichia coli is a main cause of microbial contamination in lettuce. Detection of microbial contamination is the essential key to ensure the consumption of fresh lettuce. In this investigation, the potential of Vis/NIR spectroscopy system with chemometrics analysis was probed for detecting different microbial loads (0.1, 0.2 and 0.3 ml concentration of E. coli solution) on the lettuce samples in the wavelength range of 350–1100 nm. Five different chemometrics analyses consist of soft independent modeling of class analogies (SIMCA), support vector machine (SVM), Partial least Squares Discriminant Analysis (PLS-DA) (supervised techniques), principal component analysis (PCA) and hierarchical cluster analysis (HCA) (unsupervised techniques) were used. The results proved that HCA were correctly clustered unsafe samples with 0.2 ml and 0.3 ml microbial contamination. However, some overlapping was observed between safe and unsafe samples with 0.1 ml microbial contamination. Vis/NIR spectral data with pattern recognition methods (SIMCA and SVM) can be obtained acceptable degree of classification (87.1% and 89.39% accuracy, respectively) between safe and unsafe samples. Compare with 5 different methods, the best model was recommended by PLS-DA using standard normal variate (SNV) + second derivate (D2) pre-processing methods in 6 optimal selected wavelengths (520, 670, 700, 750, 900, 970 nm) with the minimum standard error of cross validation (SECV = 0.176). Besides, a good correlation (rc = 0.989) between Vis/NIR spectral data and E. coli contamination proved the possibility of using Vis/NIR spectroscopy system to detection and evaluation of microbial contamination in lettuce.