Blog

When certain feature fields in a binary classification dataset exhibit uncertainty or data inaccuracies, it is necessary to establish a Robust Support Vector Machine (RSVM) to enhance the model’s robustness. Compared to standard support vector machines, RSVM is more effective in addressing uncertainty in the feature fields, but it also makes the model more complex, thereby increasing computational difficulty. Consequently, developing an effective solution algorithm becomes a critical challenge. The literature typically employs Second-Order Cone Programming (SOCP) to optimize RSVM. However, the efficiency of SOCP solutions significantly decreases as the number of feature fields in the dataset increases, leading to excessive computational costs, especially when handling large-scale datasets. To address this issue, we explored and implemented an alternative solution method aimed at improving computational efficiency for the model under large-scale datasets while maintaining effective handling of uncertainties. Such improvements not only enhance the speed of solving but also allow RSVM to be applied to a broader range of practical problems.

More information : https://researchoutput.ncku.edu.tw/zh/persons/jen-yen-lin