Application of Moving Average Filter for the Quantitative Analysis of the NIR Spectra

Abstract

This paper investigates the use of the moving average (MA) filter for the quantitative analysis of the near-infrared (NIR) spectra. The performance of the MA filter is mainly dependent on the right choice of the filter length. The paper also introduces a new technique for the determination of the appropriate length of the MA filter. The use of MA filters for the quantitative analysis of the NIR spectra has not been previously evaluated in the field of chemometrics. The effect of using several types of MA filters on the performance of the partial least squares regression (PLS) model has been studied and evaluated. The efficiency of the model was validated using different mixtures composing of glucose, urea and triacetin dissolved in a phosphate buffer solution. The results demonstrate that the use of a small filter length reduces the signal to noise ratio of the spectra, while the use of a larger filter length introduces a distortion and some information gets lost. In this paper, the exponential MA (EWMA) filter has been used specifically for the quantitative analysis of the NIR spectra for the first time. Its performance has been evaluated and compared to other types of MA and Savitzky- Golay filters. The results show that the standard error of prediction decreases from 35.6 mg/dL when 8 PLS factors are used for the PLS model to 22.4 mg/dL when 9 PLS factors are used for the EWMA−PLS model.