An entropy based approach to find the best combination of the base classifiers in ensemble classifiers based on stack generalization

In recent years, there has been an increasing interest in the area of multiple classifier system. The major objective in multi classifier system is to fuse a set of base classifiers in such a way the final output be more accurate than each base classifier. So far, different methods has been suggested for fusing the base classifiers. Nevertheless, selecting the base classifiers is usually performed manually. The accuracy of the ensemble classifiers is severely depend on the diversity among the base classifiers. Numerous studies have investigated the diversity measures for classifiers. There are wide range of learners that could be candidate for base classifiers, consequently, the optimum selection of the base classifiers is a complex and challenging issue. In this paper, an automatic selection of base classifiers relying entropy between the classifiers has been suggested. Experimental results carried out with 4 databases of UCI repository confirm the validity of the approach in execution time and the quality of the found solution.