Feature selection SVM through Universum and its applications on text classification Feature selection SVM through Universum

The continuous emergence of digital texts makes text classification one of the key tasks. Support Vector Machine (SVM) has become a widely used classification tool due to its strong generalization ability and dependence on a few parameters. However, SVM was not originally designed to determine relevant features. This research focuses on applying SVM and Universum learning to text classification, exploring their effects on processing unlabeled data, enhancing generalization capabilities, and solving the problem of feature selection. By introducing the Universum set, we construct an extended data set and incorporate the concept of Universum embedding. We propose a Feature Selection Universum Support Vector Machine (FSUSVM). This model introduces constraints on the prediction boundaries of the Universum set on this extended dataset to ensure its robust performance in terms of feature selection. Specifically, by incorporating constraints on the prediction boundaries of the Universum set into the existing SVM model, we aim to optimize the model's accuracy and feature selection performance in text classification tasks. F Finally, we substantiated the effectiveness of FSUSVM through numerical experiments conducted on text data. Additionally, we evaluated FSUSVM on image data, yielding positive results.