Optimizing Weights in Combining Classifiers in Natural Language Learning

S.-B. Park and H. Yoon (Korea)


Computational Linguistics, Committee machine, Optimiz ing weights, and Genetic algorithm


Many machine learning algorithms have their own idiosyn crasy in their generalization, even though they have been successful in various tasks. Therefore, in many real-world tasks, a committee of several diverse classifiers outper forms any single committee member. However, it is yet an open problem how to combine them in order to achieve high performance. This paper proposes a novel method based on genetic algorithms for combining multiple classi fiers. The experimental results on natural language learning show that the proposed method is plausible for combining classifiers. The combination of na¨ıve Bayes classifier, deci sion trees, and memory-based learning achieves on average 90.14% of accuracy for compound noun decomposition of Korean, while the base classifiers give 73.17%, 82.28%, and 86.26% of accuracy respectively.

Important Links:

Go Back