Modeling a Classifier for Solving Safety-Critical Binary Classification Tasks

This paper introduces a novel machine learning approach for performing binary decision-making tasks under uncertainty. Reducing the regression test effort of safety-critical black box systems is a safety-critical task as system failures would remain undetected if corresponding failing test cases are not executed. The uncertainty of potentially undetected system failures persists due to the lack of implementation knowledge of black-box systems. We refer to executing test cases as a costly labeling process due to required special test equipment and testing time. However, we model a binary classifier for selecting test cases. Accordingly, only fault revealing test cases should be selected and thus executed in order to reduce the overall cost of the regression test effort. On the one side, the classifier's specificity has to be maximized. On the other side, the classifier's sensitivity has to meet a specific quality-level since the number of undetected system failures should be limited. We will show in an industrial case study the benefits of our classifier where we reduce the regression test effort of safety-critical systems. The experimental results indicate that our implemented classifier outperforms other machine learning approaches in terms of sensitivity.