Feature selection for label distribution learning using Dempster-Shafer evidence theory

• DOI: 10.1007/s10489-024-05879-z
• Publication Date: 2025-01-03T08:18:51Z
• Authors: Zhengwei Zhao; Rongrong Wang; Wei Pang; Zhaowen Li
• Keywords: Dempster–Shafer theory; Feature (linguistics); Feature vector

In the contemporary epoch of massive data, fuzziness labels and high dimensionality feature space are prevalent characteristics data. As a mathematical methodology for managing uncertainty, Dempster-Shafer evidence theory has found widespread applications in artificial intelligence, pattern recognition, decision analysis. However, it not garnered adequate attention label distribution learning (LDL). This paper studies selection LDL using theory. First, distance maps given, respectively. Furthermore, tunable parameter to regulate proximity level features or is implemented. Then, $$\alpha $$ -upper -lower approximations data put forward. Subsequently, alleviate influence uncertainty on classification performance, robust evaluation measures namely, "belief map" "plausibility defined, they based approximations. Next, algorithms utilizing belief plausibility specially designed. Finally, experimental results statistical analyses demonstrate that defined can effectively measure indeterminacy designed outperform five existing regarding performance.