An ensemble pruning method considering classifiers' interaction based on information theory for facial expression recognition
Yiqing Wu,Danyang Li,Xing Chen,Yumei Tang,Shisong Huang
DOI: https://doi.org/10.1007/s00530-023-01227-2
IF: 3.9
2024-01-29
Multimedia Systems
Abstract:Ensemble learning combines all generated base learners for better generalization performance, but weak and redundant classifiers reduce the classification system's performance, so the researchers proposed ensemble pruning. Existing ensemble pruning methods usually utilize the classifiers' diversity and ability to select optimal classifier sequences. However, they ignore that the interaction between two weak or redundant classifiers can improve performance. In this paper, we focus on the interaction between classifiers and propose a new ensemble pruning method CCIEP (An Ensemble Pruning Method Considering Classifiers' Interaction). We apply CCIEP mainly to facial expression recognition. CCIEP consists of two parts: first, we use symmetric uncertainty as the ranking metric, and then we use symmetric uncertainty to perform ranking-based pruning of the classifier pool to ensure the ensemble performance of a subset of classifiers. Second, we use the interaction information to explore the binding relationships between classifiers. Then, we add classifiers with interactions to the selected subset of classifiers, which can optimize the system performance by adding interacting classifiers. Experimental results demonstrate that the method outperforms some state-of-the-art ensemble pruning methods on five classical face expression data sets and 10 UCI data sets.
computer science, information systems, theory & methods