Regression based on neural incremental attribute learning with correlation-based feature ordering

Incremental Attribute Learning (IAL) gradually trains features in one or more size, which can be used to solve regression problems. Previous studies showed that feature ordering is crucial to IAL, and features should be sorted by some criteria. This study proposed two new feature ordering methods based on feature´s group correlation and individual correlation for different situations. Experimental results show that grouped correlation-based feature ordering approach can exhibit better performance than others based on IAL neural networks in regression. Moreover, the performance of this approach is more stable than individual correlation-based approaches and some other approaches.