Random Forest as a non-parametric algorithm for near-infrared (NIR) spectroscopic discrimination for geographical origin of agricultural samples

Recently, agricultural products from diverse geographical origins are widely available in markets due to increased international trading via free trade agreement (FTA). There-fore, there is a demand for analytical methods to correctly identify their geographical origins to ensure public confidence in fair evaluation of product value. The use of chromato-graphic methods and/or DNA analysis enables the identi-fication of geographical origins; however, these are slow and requiring extensive sample preparation.1-6 When number of target samples is large and efficient screening of them is necessary, a fast and non-destructive analytical method such as near-infrared (NIR) and Raman spectroscopy is adequate, although the subsequent accuracy of identification would degrade. Since agricultural products are complex in composition