Feature Selection using Tabu Search with Learning Memory: Learning Tabu Search

Lucien Mousin (1, 2) , Laetitia Jourdan (1, 2) , Marie-Eléonore Marmion (1, 2) , Clarisse Dhaenens (1, 2)

1 Université de Lille, Sciences et Technologies
2 DOLPHIN - Parallel Cooperative Multi-criteria Optimization

Feature selection in classification can be modeled as a com-binatorial optimization problem. One of the main particularities of this problem is the large amount of time that may be needed to evaluate the quality of a subset of features. In this paper, we propose to solve this problem with a tabu search algorithm integrating a learning mechanism. To do so, we adapt to the feature selection problem, a learning tabu search algorithm originally designed for a railway network problem in which the evaluation of a solution is time-consuming. Experiments are conducted and show the benefit of using a learning mechanism to solve hard instances of the literature.

Domaines: Recherche opérationnelle [math.OC]

Lucien Mousin, Laetitia Jourdan, Marie-Eléonore Marmion, Clarisse Dhaenens. Feature Selection using Tabu Search with Learning Memory: Learning Tabu Search. Learning and Intelligent OptimizatioN Conference LION 10, May 2016, Ischia Island (Napoli), Italy. ⟨hal-01370396⟩