Designing a framework to gain insight into the automation of defect management at NLR

In this research a framework for decision-making in the defect management process is designed and the amount of data that is needed for reliable results. Defect management is the process that starts when a defect occurs and is about the rectification of said defect. In this research two options are considered: immediate rectification and deferral of this rectification. 10 scenarios are constructed and experts are asked to decide for those scenarios what the best decision is and on which variables their decision is based. Their decision logic is used to develop a decision tree that can decide for all scenarios in the scope of this research what the decision would be. This decision tree is used to decide in randomly generated scenarios which maintenance rectification option is preferred. On this data set, four classifiers try to learn back the decision tree. Using stratified 10-fold cross-validation the random forest classifier and gradient boosting classifier can with 500 scenarios decide accurately in 99.6% of the scenarios. However, the importance of the variables when doing so does not align with our expectations based on the decision tree. With 5000 scenarios, the variable importance does align better with our expectations.