Resolve Overlapping Voltammetric Peaks by Artificial Neural Networks with Maximum Likelihood Principal Component Analysis
Ling Gao, Xiaoping Li, Shouxin Ren · 2008
The multilayer feedforward networks based on the back propagation with maximum likelihood principal component analysis (MLPCA-BP-MLFN) preprocessor were developed to analyze overlapping Osteryoung square wave voltammograms. The principal component analysis back propagation multilayer feed forward networks (PCA-BP-MLFN) and the Kernel Partial Least Squares (KPLS) method were also applied in this paper for comparison. Three programs called PKPLS, PPCABPMLFN and PMLPCABPMLFN were designed to perform the calculations. A comparative study of the prediction capabilities of the three approaches showed the three methods provided satisfactory results. MLPCA-BP-MLFN is a valuable tool in solving the local minimum problem and improving the convergence rate. Comparing with KPLS and PCA-BP-MLFN, MLPCA-BP-MLFN was showed to be improved significantly in the case.