@misc{15061,
  keywords = {Machine learning, Multi-Objective Search, Configuration, Product Line, Rule Mining, Interacting Products},
  author = {Safdar Safdar and Tao Yue and Shaukat Ali and Hong Lu},
  title = {Employing Multi-Objective Search and Machine Learning to Mine Cross Product Line Rules: A Technical Report},
  abstract = {Modern systems are being developed by integrating multiple products within/across product lines that communicate with each other through information networks. Runtime behaviors of such systems are related to product configurations and information networks. Cost-effectively supporting Product Line Engineering (PLE) of such systems is challenging mainly because of lacking the support of automation of the configuration process. Capturing rules is the key for automating the configuration process in PLE. However, there does not exist explicitly-specified rules constraining configurable parameter values of such products and product lines. Manually specifying such rules is tedious and time-consuming. To address this challenge, in this paper, we present an improved version (named as SBRM+) of our previously proposed Search-based Rule Mining (SBRM) approach. SBRM+ incorporates two machine learning algorithms (i.e., C4.5 and PART) and two multi-objective search algorithms (i.e., NSGA-II and NSGA-III), employs a clustering algorithm (i.e., k-means) for classifying rules as high or low confidence rules, which are used for defining three objectives to guide the search. To evaluate SBRM+ (i.e., SBRM+NSGA-II-C45, SBRM+NSGA-III-C45, SBRM+NSGA-II-PART, and SBRM+NSGA-III-PART), we performed two case studies (Cisco and Jitsi) and conducted three types of analyses of results: difference analysis, correlation analysis, and trend analysis. Results of the analyses show that all the SBRM+ approaches performed significantly better than two Random Search-based approaches (RBRM+-C45 and RBRM+-PART) in terms of fitness values, six quality indicators, and 17 machine learning quality measurements (MLQMs). As compared to RBRM+ approaches, SBRM+ approaches have improved the quality of rules based on MLQMs up to 27\% for the Cisco case study and 28\% for the Jitsi case study.},
  year = {2018},
  pages = {1-59},
  month = {05/2018},
  publisher = {Simula Research Laboratory},
  address = {Oslo, Norway},
  url = {https://www.simula.no/sites/default/files/publications/files/employing_multi-objective_search_and_machine_learning_to_mine_cross_product_line_rules.pdf},
  note = {This work was supported by the Zen-Configurator project funded by the Research Council of Norway (Grant No. 240024/F20) under the category of Young Research Talents of the FRIPO funding scheme. Tao Yue and Shaukat Ali are also supported by the Co-evolver project funded by the Research Council of Norway (grant no. 286898/LIS) under the category of Researcher Projects of the FRIPO funding scheme.},
}