Racing-Based Parameter Selection for Optimizing Multi-Objective Classifiers

One of the key problems in data analytics is classification, i.e. classifying new data into known categories by learning from previous data. An example for this would be the detection of customer solvency [1]. Often, the classifiers have many parameters and it is of great importance to configure them accurately in order to obtain the best performance. So far, there are a number of methods for tuning these parameters, even in the presence of multiple performance criteria [2]. However, those methods often rely on outdated algorithms. Thus, the aim of this project is two-fold:


  1. Explore the application of state-of-the-art evolutionary multi-objective algorithms, such as SMS-EMOA [3], NSGA-III [4], IBEA, etc; and
  2. Combine the multi-objective optimizer with a racing procedure [5] in order to make the stochastic evaluation more efficient.

Literature / References:


  • [1] Daskalaki, S., Kopanas, I., Goudara, M., Avouris, N. (2003). Data mining for decision support on customer insolvency in telecommunications business. European Journal of Operational Research, 145(2), 239-255. [PDF]

  • [2] Müssel, C., Lausser, L., Maucher, M., & Kestler, H. (2012). Multi-Objective Parameter Selection for Classifiers. Journal of Statistical Software, 46(5), 1-27. [PDF]

  • [3] N. Beume, B. Naujoks, and M. Emmerich. SMS-EMOA: Multiobjective selection based on dominated hypervolume. European Journal of Operational Research, 181(3), 1653-1669. [PDF]

  • [4] Deb, K., & Jain, H. (2014). An evolutionary many-objective optimization algorithm using reference-point-based nondominated sorting approach, part I: Solving problems with box constraints. IEEE Transactions on Evolutionary Computation, 18(4), 577-601. [PDF]

  • [5] M. Birattari. The race package for R: Racing methods for the selection of the best. Technical Report TR/IRIDIA/2003-037, IRIDIA, Université Libre de Bruxelles, Belgium, 2003. [software]