Comparison of Different Pre-processing Techniques towards Discrimination of Turmeric Powders using Near-infrared Spectra and Exploratory Data Analysis

Pre-processing of near-infrared (NIR) spectroscopic data is the initial step for feature extraction, playing an significant role for the establishment of consistent results in both quantitative as well as qualitative analysis. In this paper different pre-processing techniques and their combination is compared for real dataset of NIR spectral data of turmeric powders in the transmittance mode. The exploratory data analysis was done by principal component analysis (PCA) which visually display the clustering between different groups of data. The improvement of the clustering has been quantitatively found through a mathematical expression called separability index (SI). Finally the best pre-processing techniques are obtained from the comparison graph between SI for different pre-processing techniques.