Abstract | ||
---|---|---|
Feature selection (FS) has proven its importance as a preprocessing for improving classification performance. The success of FS methods depends on extracting all the possible relations among features to estimate their informative amount well. Fuzzy information measures are powerful solutions that extract the different feature relations without information loss. However, estimating fuzzy information measures consumes high resources such as space and time. To reduce the high cost of these resources, this paper proposes a novel method to generate FS based on fuzzy information measures using descriptive statistics data (DS) instead of the original data (OD). The main assumption behind this is that the descriptive statistics of features can hold the same relations as the original features. Over 15 benchmark datasets, the effectiveness of using DS has been evaluated on five FS methods according to the classification performance and feature selection cost. |
Year | DOI | Venue |
---|---|---|
2022 | 10.1007/978-3-031-10989-8_7 | Knowledge Science, Engineering and Management |
Keywords | DocType | ISSN |
Feature selection, Fuzzy information measures, Fuzzy sets, Descriptive statistics, Classification systems | Conference | 0302-9743 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Salem Omar A. M. | 1 | 0 | 0.34 |
Liu Haowen | 2 | 0 | 0.34 |
Feng Liu | 3 | 85 | 17.02 |
Yi-ping Phoebe Chen | 4 | 1060 | 128.42 |
Xi Chen | 5 | 333 | 70.76 |