Methods for outlier detection in metrological studies

Abstract

The article addresses the issue of outliers in metrological measurements, which can significantly distort research results and affect measurement accuracy. Outliers that substantially differ from other data points in a sample seriously threaten the reliability of metrological processes. In previous studies, the Isolation Forest model was applied to detect such outliers, demonstrating its effectiveness under certain conditions. For a deeper understanding and validation of the results, it is necessary to compare this approach with traditional robust methods, such as the Interquartile Range (IQR) and Median Absolute Deviation (MAD), already widely used in metrology. This work compares the mentioned outlier detection methods with the Isolation Forest model. Special attention is given to the impact of outliers on data distribution and each method's ability to impact mitigation, enhancing reliability. The study encompasses an analysis of the characteristics of the method for the identification of strengths and weaknesses in the context of real metrological tasks.