원문정보
초록
영어
The existing classifiers for uncertain data don’t consider the dynamic cost, so this paper proposes the classification approach of the dynamic cost-sensitive decision tree for uncertain data based on the genetic algorithm (GDCDTU) , which overcomes the limitations of the stationary cost, and searches automatically the suitable cost space of every sub datasets. Firstly, this paper gives the dynamic cost- sensitive learning thought, and disposes the continuous and discrete attributes for uncertain data by the probabilistic cardinality. Secondly, we give the selection methods for the splitting attributes and the construction process for cost-sensitive decision tree, and the interval number for describing dynamic cost is coded by its centre and radius. At last, the dynamic cost-sensitive decision tree for uncertain data is structured, which uses the genetic algorithm as the optimal misclassification cost searching way, and the optimum cost is got by the hybridization, the mutation, the selection. The experiments using both artificial and real data sets show that, compared to the other decision tree classification algorithms for uncertain data, GDCDTU has higher classification accuracy and performance, and the total expenditure is lower.
목차
1. Introduction
2. Dynamic Cost-sensitive Learning
3. Description for Uncertain Data
3.1 Uncertain Discrete Attribute
3.2. Uncertain Continuous Attribute
4. Dynamic Cost-sensitive Decision Tree for Mining Uncertain Data Based on the Genetic Algorithm
4.1. Chromosome Coding
4.2. Selection of Splitting Attribute
4.3. Structure of Cost-sensitive Decision Tree for Uncertain Data
4.4. Fitness Function
4.5 Classification Algorithm of Cost-sensitive Decision Tree for Uncertain Data
5. Simulation Experiment
6. Conclusion
Acknowledgements
References