Designing multiple classifier combinations a survey

• Main Authors: Husin, Abdullah, Ku-Mahamud, Ku Ruhana
• Format: Article
• Language: English
• Published: Little Lion Scientific 2019
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://repo.uum.edu.my/27859/1/JTAIT%2097%2020%202019%202386%202405.pdf; http://repo.uum.edu.my/27859/; http://www.jatit.org/volumes/ninetyseven20.php
• Institution: Universiti Utara Malaysia

Classification accuracy can be improved through multiple classifier approach. It has been proven that multiple classifier combinations can successfully obtain better classification accuracy than using a single classifier. There are two main problems in designing a multiple classifier combination which are determining the classifier ensemble and combiner construction. This paper reviews approaches in constructing the classifier ensemble and combiner. For each approach, methods have been reviewed and their advantages and disadvantages have been highlighted. A random strategy and majority voting are the most commonly used to construct the ensemble and combiner, respectively. The results presented in this review are expected to be a road map in designing multiple classifier combinations.