Open AccessOptimising ensemble combination based on maximisation of diversity
Author(s) -
Mao Shasha,
Lin Weisi,
Chen Jiawei,
Xiong Lin
Publication year - 2017
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
eISSN - 1350-911X
pISSN - 0013-5194
DOI - 10.1049/el.2017.0795
Subject(s) - ensemble learning , diversity (politics) , constraint (computer aided design) , computer science , exploit , machine learning , artificial intelligence , ensemble forecasting , algorithm , mathematics , geometry , computer security , sociology , anthropology
Balancing diversity and accuracy of individuals is crucial for improving the performance of an ensemble system, since they are two important but incompatible factors for ensemble learning. When multiple individuals are combined with the corresponding weights, the diversity should be dominated by individuals and their weights, whereas the weights are normally ignored in the analysis of diversity in most research. Inspired by this, the authors propose a novel ensemble method which seeks an optimal combination to maximise diversity and accuracy of weighted individuals with the constraint on the minimal ensemble error. Furthermore, a new expression is given based on the generated individuals and their weights to exploit the diversity of an ensemble. Experimental results illustrate that the proposed method outperforms relevant existing methods.