Outlier detection in UV/Vis spectrophotometric data

UV/Vis spectrophotometers have been used to monitor water quality since the early 2000s. Calibration of these devices requires sampling campaigns to elaborate relations between recorded spectra and measured concentrations. In order to build robust calibration data sets, several spectra must be recorded per sample. This study compares two approaches – principal component analysis and data depth theory – to identify outliers and select the most representative spectrum (MRS) among the repetitively recorded spectra. Detection of samples that contain outliers is consistent between the methods in more than 70% of the samples. Identification of spectra as outliers is consistent in more than 95% of the cases. The identification of MRS differs depending on the approach used. In their current form, both of the proposed approaches can be used for outlier detection and identification. Further studies are suggested to combine the methods and develop an automated ranking and sorting system.