Savitzky-Golay Parameter Optimization by using Linear Discriminant Analysis for FTIR Spectra

The Savitzky-Golay (SG) smoothing technique has been successfully applied to filter the noise in spectra or chromatograms. In this research, SG smoothing was applied as a pre-processing method to analyze Fourier Transform Infrared (FTIR) spectra of chilli plants infected by Pepper Yellow Leaf Curl Virus (PYLCV) and PYLCV-undetected plants. SG smoothing has two parameters that can be optimized to achieve the best result, namely the polynomial order and the window length. For the former, orders of zero, two, four, six and eight whilst the latter used lengths from the polynomial order + 1 to 59. Linear Discriminant Analysis (LDA) was used to optimize the parameters. The results showed that the best LDA classification result was achieved using the zeroth and second order polynomials. For the zeroth order, a 100% classification result was achieved by window lengths in the range nine to twenty-five, while the second order polynomial window lengths to achieve the same results were from twenty-nine to forty-one. From the two polynomial orders, the mean squared error (MSE) of the SG smoothed, and the original signal was calculated. From that process, the zeroth order SG smoothing curve with a window length of nine produced the best parameter combination to classify the samples.