Combination of Transferable Classification With Multisource Domain Adaptation Based on Evidential Reasoning

In applications of domain adaptation, there may exist multiple source domains, which can provide more or less complementary knowledge for pattern classification in the target domain. order to improve accuracy, a decision-level combination method is proposed multisource adaptation based on evidential reasoning. The results obtained from different domains usually have reliabilities/weights, are calculated according consistency. Therefore, discounted by corresponding weights under belief functions framework, and then, Dempster's rule employed combine these results. reduce errors, neighborhood-based cautious decision-making developed make class decision depending result. object assigned singleton if its neighborhoods be (almost) correctly classified. Otherwise, it cautiously committed disjunction several possible classes. By doing this, we well characterize partial imprecision error risk as well. A unified utility value defined here reflect benefit such classification. This achieve maximum because considered better than an error. Several real data sets used test performance method, experimental show that our new efficiently accuracy with respect other related methods.