Discounted Belief Decision Tree for Uncertainty Data from Unreliable Source
Juntao Zou, Xuefeng Yan, Yong Zhou · 2019
In real world applications, data are often uncertainty and imperfect. The framework of Evidence theory is widely used to represent uncertainty. However, Dempster-Shafer evidence theory (DST) requires all sources of evidence are reliable. When combining evidence from unreliable sources, it will produce a counterintuitive result. A Discounted Belief Decision tree (DBDT) has been proposed a belief decision tree based on evidence discount to deal with the classification of uncertain data from unreliable sources. The discounting factor is calculated by the disagreement degree of the evidences to discount the evidences from unreliable sources. The transfer belief model (TBM) is used to transfer the belief function of discounted evidences to pignistic probability which is be used to calculate the information entropy. Experimental results show that DBT is a practical solution to solve the problem of uncertain data classification with unreliable source and it performs better than the existing algorithm.