Design of Heuristic Ensemble Classifiers with High Reliability

Sometimes, the reliability in decision of a classifier is more important than its recognition rate. Military and security applications are clear examples to show the importance of this measure. For example, the inability of an automatic targets recognition system to distinguish all types of military planes increases its error rate but the decision of this system for recognition of military targets should be accompanied with maximum reliability and never should be considered a civilian as a military target. This paper presents an ensemble classifier with high reliability by using multi-objective heuristic methods. Moreover, ensemble size and error rate have been minimized. Multi-Objective Particle Swarm Optimization Algorithm and Multi-Objective Inclined Planes Optimization Algorithm are the multi-objective heuristic methods which are used in this paper. The recent method is applied to design ensemble classifiers for the first time. Due to the ability of multi-objective heuristic methods in presentation of the Pareto front, it's possible to create various and user-defined conditions; conditions in which the importance of each factor (ensemble size, error rate and reliability) can be strengthened and weakened.