Comparison of three different classification methods performance for the determination of biofuel quality by means of NIR spectroscopy

Considering the European goal to promote clean, sustainable, and affordable energies, solid biofuels could play an important role in increasing the share of renewable energies. Given the huge variability of solid biofuel characteristics and in order to determine their best application, it is important to assess their quality before the intended use. Technical standards EN ISO 17225 divide the solid biofuels in several quality classes on the basis of chemical‐physical parameters and qualitative features (ie, origin and source of the material). These last are quite hard to be determined in densified solid biofuels like pellet and briquettes using conventional lab analysis, and near‐infrared spectroscopy could represent a rapid and economic method to overcome this problem. To this aim, three near‐infrared spectral datasets have been investigated to evaluate the possibility to get information about qualitative features of the densified solid biofuels and in detail: (a) discrimination between treated/virgin wood; (b) discrimination between bark/wood; and (c) discrimination between herbaceous/woody biomass. Three different classification methods have been taken into consideration—support vector machines (SVM), partial least squares discriminant analysis (PLS‐DA), and principal component analysis linear discriminant analysis (PCA‐LDA)—and the classification performance were compared. All the methods were carefully validated using two different new modifications of repeated double cross‐validation.