NIR spectrometry-based milk fat content classification using bagging ensembles

The short-wave near-infrared spectroscopy at 540–910 nm region is investigated for non-destructive multivariate analysis of fat content for packaged milk in four categories: Double toned, full cream, standard and toned. Visible nearinfrared spectrometry is used in the discrimination (classification) of milk fat content while red, green, blue component spectra are recorded for each sample under each aforesaid category. Features are extracted considering the highest 30 peaks of each spectra-red, green, blue component. Ensembles of classifier based on bagging strategy is employed here for the classification of samples. Two types of base classifier used here namely, support vector machine and multi-layer perceptron network. Result shows that support vector machine supersede multi-layer perceptron as individual learner in terms of classification accuracy. Single classifier performance is also compared with their native bagging-based ensemble. It is found that the bagging-based ensemble of classifier exhibits promising result in improving the prediction accuracy.