Feature Ordering for Neural Incremental Attribute Learning Based on Fisher´s Linear Discriminant

Incremental attribute learning (IAL) often gradually imports and trains pattern features in one or more size, which makes feature ordering become a novel preprocessing work in IAL process. In previous studies, the calculation of feature ordering is often Based on feature´s single contribution to outputs, which is similar to wrapper methods in feature selection. However, such a process is time-consuming. In this paper, a new approach for feature ordering is presented, where feature ordering is ranked by Fisher Score, a metric derived by Fisher´s Linear Discriminant (FLD). Based on neural network IAL model, experimental results verified that feature ordering derived by Fisher Score can not only save time, but also obtain the best classification rate compared with those in previous studies.