LAUSR

Stability analysis of heuristic ensemble classifiers, which are designed by using heuristic methods, is a significant topic due to the stochastic nature of heuristic algorithms. Considering the importance of this issue, the novelty of this paper is stability analysis of heuristic ensemble classifiers. So, in this paper, at first, two multi-objective heuristic ensemble classifiers by using a new multi-objective heuristic approach called multi-objective inclined planes optimization (MOIPO) algorithm and a conventional one called multi-objective particle swarm optimization (MOPSO) algorithm are designed and then, two-level factorial designs, as a statistical approach, are applied to investigate the stability of the best ensemble classifier from two designed ensemble classifiers for the first time; for this purpose, the effects of three structural parameters of winner algorithm i.e. inflation rate, leader selection pressure and deletion selection pressure on the performance of designed heuristic ensemble classifier for three datasets as a representative of simple data, overlapped data and data with huge number of features are investigated. Extensive experimental and comparative results on different kinds of benchmarks with nonlinear, overlapping class boundaries and different feature space dimensions not only show the supremacy of MOIPO for designing ensemble classifiers but also the important parameters and important interactions for each objective function.