A novel approach to classify edible oil using multiple classifier fusion based on spectral data

Abstract

The paper investigates the discrimination ability of the short-wave near-infrared spectroscopy for the classification of various edible oils used in India. Spectral characterization of mustard, olive, rice bran, sunflower and soybean oils are recorded by a portable near infrared (NIR) spectrometer in the range of 250–925 nm of wavelength. Due to high dimensionality problem, principal component analysis (PCA) is used here for extracting only the significant features from the near infrared spectra. Hidden clusters in the dataset are identified by applying the principal component analysis (PCA). For classification purpose multilayer perceptron (MLP) and support vector machine (SVM) are used in this work. Moreover, classifier fusion strategy using Dempster Shafer Evidence Theory is implemented to increase the prediction accuracy. The results clearly indicate the efficacy of multiple classifier fusion technique when applied to near infrared spectrometry data for edible oil classification.