نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مهندسی کامپیوتر- دانشکده فنی و مهندسی ـ دانشگاه تربت حیدریه
2 دانشگاه سجاد
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
The challenge of data classification by insufficient labeled data can be solved by domain adaptation techniques and leveraging external knowledge. However, most of these techniques lose robustness in noisy environments where the labels and features become corrupted. Aiming to model the indiscernibility and vagueness in domain adaptation, the present paper introduces a twin model for domain adaptation that combines the quarter-sphere support vector machine data description (QS-SVM) with a new fuzzy rough set-based weighting approach. The proposed model learns two small hyperspheres per domain regarding binary classification by solving two linear equations rather than one Quadratic Programming Problem (QPP), unlike standard QS-SVM. Consequently, the time complexity is reduced by this strategy. The Benefit of the fuzzy rough set is that only the high-confidence samples influence the adaptation and classification results of the hyperspheres. The strength of the proposed model is that after constructing and training the source domain classifiers, accessibility to the source domain data is not required, and the existence of only the source domain hyperspheres is sufficient. Also, the proposed fuzzy rough set-based sample weighting method ensures that the minority classes that are often underrepresented in the dataset are not overlooked when constructing the model. The effectiveness of the proposed model has been compared to the state-of-the-art methods on fifteen tasks taken from two benchmark datasets. The experimental results demonstrate the superiority of the proposed model over state-of-the-art ones in terms of classification accuracy and computational time. Besides, the noise analysis proves the robustness of the proposed model.
کلیدواژهها [English]
)