Rapid classification of peanut varieties for their processing into peanut butters based on near‐infrared spectroscopy combined with machine learning

Peanut classification based on processing purposes is becoming mainstream. In order to speed up the classification procedure, near-infrared (NIR) spectroscopy for classifying peanut varieties for their processing into peanut butters was assessed for the first time. Peanut varieties were primarily classified by principal component analysis (PCA) combined with cluster analysis based on the structural characteristics (texture and rheology) and roast characteristics (colour and volatile compounds) of the resulting peanut butters. After the completion of spectral collection and subsequent spectral pre-treatments, the performances of classification models built by partial least squares discriminant analysis, support vector machine, and random forest were compared. PCA, variable importance, and random forest selection by filter were investigated as feature extraction methods. The sensitivity, specificity, and accuracy of the filtered cross validation and external validation models were all over 90%, while the kernel density estimation presented the acceptable distribution results of categories probabilities in the selected models. These results showed that NIR spectroscopy combined with machine learning methods is a promising approach to provide a reliable evaluation of peanuts for efficient processing.