Evaluating Machine Learning Algorithms in Representing Decision Makers in search-based PLA

Search-Based Software Engineering (SBSE) techniques have achieved satisfactory results for the optimization of Product Line Architecture (PLA) design by means of the Multi-objective Optimization Approach for PLA design (MOA4PLA) and its tool, called OPLA-Tool. However, the obtained solutions in SBSE might be rejected by the Decision Maker (DM) in some cases because many aspects of the problem cannot be mathematically modeled. Thus, in a previous work we contributed to developing an interactive module, which was incorporated in OPLA-Tool v2.0, to support the DM’s preferences during the optimization process. This module includes a Machine Learning Model (ML) based on Multilayer Perceptron (MLP) to prevent human fatigue, caused mainly by the excessive number of inter-actions demanded by evolutionary algorithms. The performance of ML algorithms varies according to the problem to be solved. Thus, an unexplored gap in the previous work is the analysis of the ML model induced by different algorithms. In this context, this work aims at evaluating different ML algorithms in the interactive module of OPLA-Tool v2.0. We conducted an exploratory study with the DM participation using different ML algorithms. The obtained results increase the state of the art, as other ML algorithms outperform the MLP used in our previous work in terms of processing time and accuracy.