Corrigendum to “An improved sensor calibration with anomaly detection and removal” [Sens. Actuators, B 307 15 March (2020) 127428](S0925400519316272)(10.1016/j.snb.2019.127428)

Xinwei Fang*, Iain Bate

*Corresponding author for this work

Research output: Contribution to journalComment/debatepeer-review


The authors regret for the following errors and obfuscations The Equation (8) It is currently read:Completeness=#TruePositives#TruePositives+#TrueNegatives It should read:Completeness=#TruePositives#TruePositives+#FalseNegatives In the Section 4.4.2 Evaluation on Beijing dataset The authors would like to clarify two points. 1) It was incorrect to say we could not reproduce the work in [1]. We could reproduce the method but got slightly different results. An investigation has determined that our paper used different variables to train the Gaussian Process (GP) used for sensor calibration to those used in [1], fully explaining why our GP calibration was less successful than the Smith et al. GP. 2) Our paper also does not clearly state the data used to set up the GP, in particular that some of the data used to train our GP comes from expensive reference grade instruments in contrast to Smith et al. who used only data from low-cost sensors. Using different input data does not affect the validity of the results or the insight gained, however clearly different trends may be seen if the same input data is used as Smith used. The authors would like to apologise for any inconvenience caused.

Original languageEnglish
Article number128239
JournalSensors and Actuators, B: Chemical
Publication statusPublished - 15 Aug 2020

Cite this