B0348
Title: Vine copula based classifiers
Authors: Ozge Sahin - Delft University of Technology (Netherlands) [presenting]
Harry Joe - University of British Columbia (Canada)
Abstract: An innovative approach is presented for classification tasks that combine feature selection and vine copulas fitted to the distribution of features in each class. The power of vine copulas is leveraged to capture complex dependencies among features and that of the proposed feature selection methods to enhance the accuracy and robustness of the classifiers. Categorical prediction intervals are introduced to summarize the classifier's performance beyond point predictions. Through extensive experiments on real data, the superior performance of the approaches is demonstrated, compared to traditional discriminative methods and random forests when features have different dependent structures for different classes. The research contributes significantly to the classification field by offering a powerful framework that combines feature selection, vine copulas, and Bayesian inference for accurate and interpretable classification in high-dimensional data analysis.