相互信息
计算机科学
人工智能
球(数学)
特征选择
粒度计算
模糊逻辑
模式识别(心理学)
模糊集
数学
粗集
数学分析
标识
DOI:10.1145/3697467.3697656
摘要
Feature selection, stands as a pivotal preprocessing step in machine learning, reduces the dimensionality of data while maximizing the retention of the original categorical information within the data. Existing feature selection algorithms based on fuzzy neighborhood rough sets usually use the same fuzzy neighborhood radius for all samples when generating fuzzy equivalence relations, yet this may ignore the respective specificity of the samples, which may result in degradation of the model performance. In order to address this issue, a new multi-label feature selection algorithm is proposed by combining the fuzzy neighborhood mutual information with the granular ball model. Firstly, Considering inter-sample specificity. the samples are partitioned such that samples in the same ball use the same fuzzy neighborhood radius for generating fuzzy equivalence relations, while samples in different balls have different fuzzy neighborhood radius, and the formula for calculating the radius is given, thus preserving the specificity of the samples. Secondly, the granular ball model is used to divide the samples, in which we improve the granular ball model to make it suitable for multi-label data. Thirdly, we propose a scoring function using fuzzy neighborhood mutual information, a scoring function that fully takes into account the redundant information that still exists in the selected features and combines the correlation between features and the correlation between features and labels. Ultimately, to ascertain the efficacy of our algorithm, we undertake a comparative analysis, The experimental results show that the algorithm proposed in this paper achieves better performance than the three compared feature selection algorithms in terms of the four evaluation metrics.
科研通智能强力驱动
Strongly Powered by AbleSci AI