Multiple Testing for Outlier Detection in Space Telemetries

We propose a novel procedure for outlier detection in space telemetries, in a semi-supervised framework. As the data is functional, we reduce its dimension by considering the coefficients obtained after projecting the observations onto orthonormal bases. A multiple testing procedure based on the two-sample test is defined in order to highlight the levels of the coefficients on which the outliers appear as significantly different from the nominal data. The Local Outlier Factor is computed on the selected coefficients to highlight the outliers. This procedure for selecting the features is applied on simulated data that mimic the behavior of space telemetries and on a real telemetry and then compared with existing dimension reduction techniques.