Automated pipeline for classifying Aroclors in soil by gas chromatography/mass spectrometry using modulo compressed two-way data objects

Seven polychlorinated biphenyls (PCBs) commercial mixtures, Aroclor 1016, 1221, 1232, 1242, 1248, 1254, and 1260, were analyzed by gas chromatography/mass spectrometry (GC/MS) combined with solid phase microextraction (SPME). Three pattern recognition methods: a fuzzy rule-building expert system (FuRES), partial least-squares discriminant analysis (PLS-DA), and a fuzzy optimal associative memory (FOAM) were used to build classification models. Modulo compression was introduced for data preprocessing to extract the characteristic features and compress the data size. Baseline correction and data normalization were also applied prior to data processing. Four GC/MS data set configurations were constructed and used to evaluate the classifiers and data pretreatments including two-way modulo compressed, two-way data, one-way total ion current and one-way total mass spectrum. The results indicate that modulo compression and baseline correction methods significantly improved the performance of the classifiers which resulted in improved classification rates for FuRES, PLS-DA, and FOAM classifiers. By using two-way modulo compressed data sets, the average classification rates with FuRES, PLS-DA, and FOAM were 100±0%, 94.6±0.7%, and 96.1±0.6% for 100 bootstrapped Latin partitions of the Aroclor standards. The classifiers were validated by application to Aroclor samples extracted from soil with no parametric changes except that the calibration set of standards and validation set of soil samples were individually mean centered. The classification rates for the GC/MS modulo 35 compressed data obtained from the Aroclor soil samples with FOAM, FuRES, and PLS-DA were 100%, 96.4%, and 78.6%, respectively. Therefore, a chemometric pipeline for SPME-GC/MS data coupled with chemometric analysis was devised as a fast authentication method for different Aroclors in soil.