An application of data mining to fruit and vegetable sample identification using Gas Chromatography-Mass Spectrometry

One of the uses of Gas Chromatography-Mass Spectrometry (GC-MS) is in the detection of pesticide residues in fruit and vegetables. In a high throughput laboratory there is the potential for sample swaps or mislabelling, as once a sample has been pre-processed to be injected into the GC-MS analyser, it is no longer distinguishable by eye. Possible consequences of such mistakes can be the destruction of large amounts of actually safe produce or pesticide-contaminated produce reaching the consumer. For the purposes of food safety and traceability, it can also be extremely valuable to know the source (country of origin) of a food product. This can help uncover fraudulent attempts of trying to sell food originating from countries deemed unsafe. In this study, we use the workflow environment ADAMS to examine whether we can determine the fruit/vegetable, and the country of origin of a sample from a GC-MS chromatogram. A workflow is used to generate data sets using different data pre-processing methods, and data representations from a database of over 8000 GC-MS chromatograms, consisting of more than 100 types of fruit and vegetables from more than 120 countries. A variety of classification algorithms are evaluated using the WEKA data mining workbench. We demonstrate excellent results, both for the determination of fruit/vegetable type and for the country of origin, using a histogram of ion counts, and Classification by Regression using Random Regression Forest with PLS-transformed data.