Unlike traditional Software Product Line Engineering (PLE), the configuration in Cyber-Physical System (CPS) PLE is not confined to design time configuration only but it expands from design time to post-deployment time and even more to runtime time due to adaptive nature of CPS. Cost-effectively supporting CPS PLE, particularly enabling automation of configuration in CPS PLE is a challenge. Capturing different types of rules is the key for enabling automation of configuration in CSP PLE. For post-deployment phase, product configurations may affect the interactions among products of CPS within/across product lines in an unknown way because there usually don’t exist explicitly specified rules on these configurations. Manually specifying rules is not feasible as it’s time-consuming and the rule information is hidden in various artifacts in CPS. To cater this challenge, we proposed an iterative approach that combines the search with the machine learning techniques to mine potential rules for post-deployment configuration. We compared the quality of rules produced from search-based approach with the random search (RS) based approach. To measure the quality of rules we evaluated them based on expert opinion, real data, and machine learning quality measurements (i.e., Accuracy, Precision, Recall, and F-measure). Results show that search-based approach has produced significantly better rules as compared to RS based on all three types of evaluations.
As mentioned in the paper , both NSGA-II and Random Search as well as Hyper Volume are implemented based on jMetal .
For the experiment data: The data for experiment includes: 1) Configuration files containing the products' configuraitons and correspoinding system states. 2) Rules mined corresponding to each iteration. 3)Data for all the evaluation metrics. 4) Results of statistical analysis.
Download The experiment data related to Video Conferencing System.
 Safdar, S. A., Lu, H., Yue, T., & Ali, S. (2017, July). Mining cross product line rules with multi-objective search and machine learning. In Proceedings of the Genetic and Evolutionary Computation Conference (pp. 1319-1326). ACM.