LAUSR

Abstract In the presented study, an easy to implement workflow based on the evaluation of a quality control sample in non-targeted analysis (outlier detection and time related trend) is proposed for the first time. The novel concept was developed and demonstrated with Fourier transform-midinfrared spectroscopy using a rapeseed oil as quality control sample. Different data evaluation strategies for outlier detection were tested and compared: (i) principal component analysis (PCA), (ii) PCA combined with Hotelling's T-squared distribution and Q-residuals for data assessment as well as (iii) various outlier score-based methods. The build models were challenged by varying measurement and storage conditions to verify the applicability of the three evaluation types (i-iii) to identify these artificially induced variations as outliers. Analogous to a control chart in targeted analysis warning and action limits (numerical decision criteria) were calculated using outlier score-based methods. The best results were achieved by the four outlier score-based methods (pre-period n = 25), where 100% of the deliberately generated outliers were identified as such.