Principal Component Analysis for Explainability of Genetic Test Generation in Satellite Image Processing

Abstract

In this thesis, we propose an approach of exploiting ideas of existing post-hoc explainability, consisting of a Principal Component Analysis, to help understanding the resulting test cases of a gentic algorithm. By gaining deeper information about the algorithms decision making, we aim to optimize its configuration and to strengthen the confidence in the results. We demonstrate the effectiveness of our approach with experimental results from the genetic test approach for the Fine Guidance System of the ESA medium-class mission PLATO.