Optimization Algorithms for One-Class Classification Ensemble Pruning

One-class classification is considered as one of the most challenging topics in the contemporary machine learning. Creating Multiple Classifier Systems for this task has proven itself as a promising research direction. Here arises a problem on how to select valuable members to the committee - so far a largely unexplored area in one-class classification. Recently, a novel scheme utilizing a multi-objective ensemble pruning was proposed. It combines selecting best individual classifiers with maintaining the diversity of the committee pool. As it relies strongly on the search algorithm applied, we investigate here the performance of different methods. Five algorithms are examined - genetic algorithm, simulated annealing, tabu search and hybrid methods, combining the mentioned approaches in the form of memetic algorithms. Using compound optimization methods leads to a significant improvement over standard search methods. Experimental results carried on a number of benchmark datasets proves that careful examination of the search algorithms for one-class ensemble pruning may greatly contribute to the quality of the committee being formed.