High accuracy determination of Angelica dahurica origin based on near infrared spectroscopy and a random forest pruning algorithm

Abstract

A near infrared spectroscopy method combined with a random forest pruning algorithm based on margin optimization and principal component analysis (PCA-MORFP) was proposed to identify the origin of Angelica dahurica. One hundred and ninety-six samples of A. dahurica were collected from four original cultivation regions; their NIR diffuse reflectance spectra were measured by a custom-built near infrared spectrometer which works in the range of 900–1700 nm with a resolution (full width at half maximum [FWHM]) of 4 nm. Combinations of Savitzky–Golay smoothing, standard normal variates, and first derivative transformations were used to preprocess the spectral data. Then the PCA-MORFP classification model was constructed. Meanwhile, the was compared with other classifying approaches, including: principal component analysis-K-nearest neighbor, principal component analysis-support vector machine, and principal component analysis-random forest. Experimental results showed that the PCA-MORFP achieved the best prediction performance over other compared methods. The recognition rates of the PCA-MORFP model were up to 100% for the calibration set and 98.2% for the prediction set, respectively. The method provides a rapid and convenient detection technique for the origin identification of A. dahurica.