Robust boosting neural networks with random weights for multivariate calibration of complex samples.

摘要: Neural networks with random weights (NNRW) has been used for regression due to its excellent performance. However, NNRW is sensitive outliers and unstable some extent in dealing the real-world complex samples. To overcome these drawbacks, a new method called robust boosting (RBNNRW) proposed by integrating version of NNRW. The builds large number sub-models sequentially robustly reweighted sampling from original training set then aggregates predictions weighted median. performance RBNNRW tested three spectral datasets wheat, light gas oil diesel fuel As comparisons RBNNRW, conventional PLS, (BNNRW) have also investigated. results demonstrate that introduction greatly enhances stability accuracy Moreover, superior BNNRW particularly when exist.