Integrating Independent Component Analysis with Artificial Neural Network to Analyze Overlapping Fluorescence Spectra of Organic Pollutants

Independent component analysis (ICA) combined with Elman recurrent neural network (ERNN) regression as a hybrid approach named ICA-ERNN was proposed for the simultaneous spectrofluorimetric determination of organic pollutants. Fluorescence spectra of these compounds under study are strongly overlapped, which does not permit direct determination without prior separation by conventional spectrofluorimetry. ICA is a blind source separation (BSS) method aiming at extracting independent source variables and their corresponding concentration profiles from the observed fluorescence spectra of chemical mixtures without using any prior knowledge about the components. The proposed method combining the idea of ICA denoising with ERNN calibration provides the ability for enhancing the extraction of characteristic information and the noise removal as well as the quality of regression. The relative standard errors of prediction (RSEP) obtained for all components using ICA-ERNN, ERNN and partial least squares (PLS) were compared. Experimental results demonstrated that the ICA-ERNN method had better result than ERNN and PLS methods and was successful even when there was severe overlap of fluorescence spectra.