Abstract: Based on the local outlier factor (LOF) for detecting outlier in knowledge discovery, a localdensityratio (LDR) based setting fuzzy membership algorithm was developed. In the proposed algorithm, the fuzzy membe rships are assigned to the samples according to their neighborhood density ratios and a single parameter selection strategy is also adopted. Numerical experiments showed that the proposed algorithm possesses a good robustness for nonlinear function estimation problems with noise data. The presented algorithm is effective for setting fuzzy memberships in fuzzy support vector machine, which is importan t to deal with classification problems and nonlinear function estimation problems with noise data.

Key words: fuzzy support vector machine, local outlier factor, localdensityratio, fuzzy membership