The use of machine learning methods in classification of pumpkin seeds (Cucurbita pepo L.)

• Published in: Genetic resources and crop evolution Vol. 68; no. 7; pp. 2713 - 2726
• Main Authors: Koklu, Murat, Sarigil, Seyma, Ozbek, Osman
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.10.2021; Springer Nature B.V
• Subjects: Agriculture; Biomedical and Life Sciences; Carbohydrates; Classification; Cucurbita pepo; evolution; Learning algorithms; Life Sciences; Machine learning; Morphology; Multilayer perceptrons; Plant Genetics and Genomics; Plant Physiology; Plant Sciences; Plant Systematics/Taxonomy/Biogeography; pumpkins; regression analysis; Review; Seeds; Support vector machines; Pumpkin seed; Logistic regression; Multilayer peceptrons; Thresholding; Classification; Random forest; Support vector machine
• ISSN: 0925-9864; 1573-5109
• DOI: 10.1007/s10722-021-01226-0

Pumpkin seeds are frequently consumed as confection worldwide because of their adequate amount of protein, fat, carbohydrate, and mineral contents. This study was carried out on the two most important and quality types of pumpkin seeds, “Ürgüp Sivrisi” and “Çerçevelik”, generally grown in Ürgüp and Karacaören regions in Turkey. However, morphological measurements of 2500 pumpkin seeds of both varieties were made possible by using the gray and binary forms of threshold techniques. Considering morphological features, all the data were modeled with five different machine learning methods: Logistic Regression (LR), Multilayer Perceptrons (MLP), Support Vector Machine (SVM) and Random Forest (RF), and k-Nearest Neighbor (k-NN), which further determined the most successful method for classifying pumpkin seed varieties. However, the performances of the models were determined with the help of the 10 k-fold cross-validation method. The accuracy rates of the classifiers were obtained as LR 87.92 percent, MLP 88.52 percent, SVM 88.64 percent, RF 87.56 percent, and k-NN 87.64 percent.