Feasibility of fourier transform-near infrared spectroscopy and machine learning for high-throughput discrimination of wild clover ( <i>Trifolium)</i> species

Clover ( Trifolium ) plants with rich protein, fiber and nutrient contents provide high forage yield, essential for efficient animal production. They also offer further ecological benefits including erosion control, soil reclamation, pollinator support, and urban greening. Wild clover species possess vital genetic traits to adapt to adverse growing conditions including drought, heat stress, and diseases. Thus, precise and quick identification of clover species can significantly aid successful plant breeding and sustainable agriculture. Human-based evaluations may not always be objective, while laboratory genetic analyses require significant amount of time, labor, and cost. The present work is the first study to accurately classify wild Trifolium species by using near infrared (NIR) reflectance spectroscopy and machine learning (ML) algorithms. NIR reflectance data (4000–10,000 cm -1 ) of 146 dried-ground plant samples belonging to nine Trifolium species were used for the classification along with four data pretreatment methods and six ML algorithms. Three different data sets were utilized in the analysis: (a) the data with no dimensional reduction, (b) the data with dimensional reduction by using principal component analysis (PCA), and (c) the data with dimensional reduction by using linear discriminant analysis (LDA). The most successful result was obtained with the LDA data coupled with multiplicative scatter correction (MSC) and K-nearest neighbors (KNN) algorithm with 98% test classification accuracy. The results of this study showed that the NIR spectroscopy coupled with ML algorithms can be utilized to correctly identify the Trifolium species needed for effective plant breeding and conservation strategies.