Classification of Malaysian forest soils by anthropogenic activities based on untargeted non-volatile organic profiles using UHPLC technique and chemometric methods for forensic provenance purposes

Abstract

Forest soils are found on land that is covered by various species of fauna and flora. Consequently, forest soils possess unique and vast chemical diversity. Classification of forest lands by anthropogenic activities could contribute to forensic soil provenance analysis. This study aimed to evaluate the feasibility of ultra-high performance liquid chromatography (UHPLC) technique in discriminating forest soils by the type of anthropogenic activities via predictive modelling. Over 160 soil samples were collected from ten sites in Peninsular Malaysia. Organic fractions of the soil samples were extracted via acetonitrile and profiled via the UHPLC analytical method proposed by previous work. An isocratic elution program was applied, and UV detection was performed at 230 nm. The raw pixel-level chromatograms were carefully optimized via single-DP and ensemble-DP strategies. Then, the performance of the preprocessed sub-datasets was evaluated based on predictive capability estimated via the K-nearest neighbour (KNN) algorithm. A stratified random resampling method was deployed in preparing 100 pairs of training and testing samples. Moreover, two sets of blind samples were also used for estimating the prediction accuracy of the KNN models. The discriminative capability of the shortlisted sub-dataset was evaluated using KNN and partial least square-discriminant analysis (PLS-DA) algorithms. The KNN and PLS-DA models, respectively achieved prediction accuracy of 90 % and 84 %. In conclusion, the UHPLC-based fingerprint technique coupled with predictive modelling shows great potential in inferring the class of trace amount of forest soil.